Artificially Intimidating

•

Channel • 2 Shows

News

News

Updated Daily
Technology

Technology

Updated Semiweekly

Aug 24

Sam Altman: I was wrong about the speed -- AI Brief August 24

Good day, humans. There is a name missing from every story today. A model near the top of the coding leaderboards that no company will admit to shipping. Harvard faculty teaching a course they are not actually in. Producers using AI and not saying so, until Dr. Dre said so out loud. Polling firms that turned out to be one guy with a website. And Sam Altman, who at least put his name on being wrong. Five stories, and in every one the interesting question is who is willing to be responsible for it. Nobody Will Say Who Owns This Model TechCrunch What happened: On August 20 a model called Ox Alpha appeared on the model marketplace OpenRouter with no company name, no press release and no logo — listed under a generic “Stealth” provider label. It is free, handles just over a million tokens of context, and is aimed at coding and long-running agent work. Developers took to it fast. Four days later, still nobody has claimed it. Why it matters: If you used it, you sent your code to servers whose owner you cannot name. The two sets of terms do not agree: the model page says prompts are retained by the provider and not used for training, while OpenRouter’s broader Stealth Program agreement says user content may be collected, retained and used for training and evaluation. As The Next Web put it, a free model is winning over developers and nobody knows whose servers it runs on. Yesterday we covered Harvey swapping frontier models for cheaper open-weight ones; this is the same trade with the vendor’s name sanded off. What everyone’s saying: The fingerprinting crowd has mostly settled on Z.ai’s GLM family — a tokenizer probe that matched 95 of 95 tests, Z.ai’s exact API error strings, and video-token budgets lining up with GLM-5V-Turbo. Early guesses ran to Google Gemini and an unreleased Microsoft model before the evidence narrowed. None of it is confirmed, because confirming it would require somebody to speak. My read between the lines: Free is not generosity here, it is the purchase price of evaluation data, and it worked beautifully. Thousands of engineers pointed production workloads at an unnamed box because the sticker said zero. We spent all of last week arguing about model safety cards and provenance, and the thing that actually got adopted this week has neither. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — the whole Ox Alpha story is a live demo of what happens when capability arrives without anyone to hold accountable for it Every story today is missing a name. Your Monday report has the opposite problem — your name is on it, and you spent Sunday night assembling it. Viktor is an AI agent that lives in Slack and connects to more than 3,000 tools, and it will pull the report, refresh the dashboard, ship the code fix and run the campaign while you are asleep. Not a chatbot you interrogate all day. A coworker you hand things to. New readers get $50 off their first month. Hire Viktor → Altman Concedes the Disruption Is Running Late Digit What happened: Speaking to podcaster David Senra, Sam Altman said he had expected the world to change much faster than it did. “I thought when we got to GPT-4, which was back in 2023, that very quickly after that there was going to be much more disruption, software businesses up for grabs right away, than it turned out to be.” His explanation: “I was wrong about a few things, but one in terms of the speed — the economy just has so much inertia.” Why it matters: This is the person whose forecasts underwrite an enormous amount of enterprise budgeting, revising the schedule in public. It is also his second walk-back of the year: in May he said he was “delighted to be wrong” about an AI jobs apocalypse. If you have been budgeting against his timelines, two data points now say to add slack to the plan. What everyone’s saying: Split down the middle. One camp reads it as the rare thing you want from a CEO, which is a scorecard with a loss on it. The other camp read it as convenient repositioning — the man who set the timelines now explaining, from inside the company that set them, why the world was too slow to keep up. My read between the lines: “The economy has inertia” is a generous way to say people looked at it and did not want it yet. Inertia is a property of the object, so the sentence puts the delay on the world rather than the product. Watch which half gets revised each time this happens. The schedule moves. The destination never does. 📖 Further reading: The Tools That Just Replaced 40% of Block’s Workforce Are Free in Your Browser — if the disruption is arriving slower than advertised, the useful question is what is already sitting in your browser waiting to be used The Brief is free and it stays free. Behind the paywall is where I stop summarising and start showing the work — the deep dives, the setups I actually ran, and the full archive. If today’s five were worth ten minutes, the rest of it is there. Become a member → Harvard Will Sell You an AI Professor TechCrunch What happened: Harvard Business School launched HBS Foundry, an eight-week, $699 bootcamp for founders in which AI avatars of its own instructors — built by the startup HeyGen, matching the real faculty’s appearance and voices — give feedback on practice pitches, sales conversations and simulated board meetings. There are weekly live sessions with actual humans; the round-the-clock coaching is the avatars. Why it matters: A school whose entire product is scarcity just put its faculty’s likeness on an assembly line and priced it at $699, which is a rounding error against the tuition of the degree those same faces teach. Every university with a recognisable name is now watching to see whether this reads as generous or as cheapening. What everyone’s saying: Two camps, both reasonable. Access: unlimited pitch practice at 2am with a Harvard-shaped coach is a real thing a founder in Ohio could not previously buy at any price. Dilution: once the credential is a rendering, it is not obvious what the $699 is actually purchasing beyond a crest. My read between the lines: The number to watch is not $699, it is the marginal cost of the ten-thousandth student, which is roughly the electricity. Harvard has spent centuries manufacturing scarcity and has just built the one product where enrolment has no ceiling. Someone in that building has done the multiplication, and the faculty in the avatars should probably ask to see it. 📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn’t Agree To. — the interesting clause in a deal like this is never the price, it is what the instructors signed away about their own faces Dr. Dre Says the Fear Is a Skill Issue Variety What happened: In a New York Times interview alongside his longtime partner Jimmy Iovine, Dr. Dre said he is using AI in his own production — “as a tool to see what it would do with what I just did” — and does not see it as a threat. “The only people that see it as a threat are the people who have trouble creating.” He compared the backlash to the way people first greeted drum machines and synthesizers. Why it matters: Iovine’s companion line is the one that actually moves the industry: plenty of producers are already using these tools and will not admit it. Dre agreed — “they’re using it, they just don’t want to admit it”. When a producer of his standing says it on the record, the thing stops being a secret and starts being a credit line somebody has to negotiate. What everyone’s saying: Musicians split along a fault line that was already there. Producers raised on sampling mostly hear “another instrument,” because recombination was always the craft. Performers and songwriters — the people whose actual voices get modelled — hear “skill issue” rather differently, with likeness suits still working through the courts. My read between the lines: Dre built a career on samples, where the raw material was always someone else’s recording and the art was in what you did next. Of course a machine that recombines feels like a bigger crate. But ask Bette Midler, who had to sue Ford over a sound-alike in the 1980s and won, and you get a different answer, because she was never in the crate business. The disagreement is not about creativity. It is about whose name is in the credits, and a drum machine never had a training set. 📖 Further reading: Embracing AI as a Superpower, Not a Shortcut — Dre’s framing is exactly the tool-versus-crutch line, and it holds up better in a studio than it does in most offices Polls Are Cheap to Fake Now The Washington Times What happened: A previously unknown outfit called Median Strategies pushed fabricated poll numbers across three states this month — one of them cited by Los Angeles Mayor Karen Bass as evidence her campaign was gaining momentum — then admitted the results were invented and described the whole thing as a “social experiment” by a 21-year-old testing how far fake data would travel. Separately, an outfit calling itself The Public Sentiment Institute admitted it had simply switched votes from one candidate to another. Why it matters: The stunt is downstream of much worse research. Dartmouth work published in the Proceedings of the National Academy of Sciences found AI-generated survey responses can pass every standard quality check pollsters use, and that across seven major national polls before the 2024 election, adding as few as 10 to 52 fake responses — at roughly five cents each — would have flipped the predicted outcome. The Washington Post has since run its own audit of which outlets and poll trackers picked up the fakes (paywalled). What everyone’s saying: Pollsters keep naming the same two accelerants. Prediction markets put a direct cash payout on moving a number, and generative AI turns a plausible polling firm — website, methodology page, pre
2d ago

Everything You Build On Belongs to Somebody Else -- AI Brief August 29

Good day, humans. OpenAI supplied the models inside Cursor for nearly four years. Then SpaceX bought Cursor, and now that pipe closes on November 12. Nobody at Cursor did anything wrong; the supplier just stopped trusting the new landlord. Bill Gates picked this week to tell the New York Times that his own industry is soft-pedalling the risks. A leaked Meta memo describes a personal agent that has its own computer and keeps working after you close the app. Hugging Face shipped a robot duck on roller skates for $399. And somebody finally counted how many agent skills are unsafe to install. Four of today's five stories come down to one question: what is running on your behalf, and who gets to switch it off? OpenAI Cuts Cursor Off at the Model OpenAI What happened: OpenAI notified SpaceX that it will wind down the contract supplying OpenAI models to Cursor, the AI coding editor, with a proposed shutoff date of November 12. SpaceX closed its $60 billion all-stock purchase of Anysphere, the startup behind Cursor, on August 14 — two weeks before the notice went out. Why it matters: Cursor is one of the most widely used AI coding tools in the world, and a large share of what it does runs on models it does not own. Nobody at Cursor shipped a bad product or broke a rule. The company changed hands, and its biggest supplier decided it no longer trusted the buyer. Developers keep working through their own API keys, so this is not a ban — it is a bill moved one layer down. What everyone's saying: OpenAI framed the call around trust rather than technology, citing Twitter breaking an OpenAI contract in 2023 and Musk admitting under oath in April that xAI distilled OpenAI's data. Bloomberg reported the wind-down. Developer reaction split cleanly between people treating it as another round of Musk-versus-Altman theatre and people pointing out that they are the ones who have to do the migration. My read between the lines: OpenAI made a point of praising Cursor's team on the way out, which is what you do when you are cutting off a partner you would rather have bought. The date is the tell. November 12 is long enough to sound reasonable in a blog post and short enough to be a deadline on somebody's sprint board. 📖 Further reading: Cursor Just Stopped Being a Code Editor — what Cursor actually became once agents moved in, and why the model underneath it was always the leverage Today's brief is full of software working while nobody is watching, and one tool that just got eleven weeks' notice. Viktor is the version that simply turns up to work. It lives in your Slack or Teams, connects to over 3,000 tools, and comes back with finished reports, dashboards, code and campaigns. Not a chatbot you have to prompt. A coworker you delegate to. New readers get $50 off their first month. Hire Viktor → Bill Gates Breaks Ranks on AI Risk The New York Times What happened: In an hourlong interview, Gates told the New York Times that the AI industry is downplaying risks he believes are real. “I don't like bringing bad news to people, and I don't like saying that innovation may be a net negative,” he said. “But that's where we are.” Why it matters: Gates has spent fifty years as the technology industry's most reliable optimist, which makes him an awkward person to dismiss. In the same interview he named the three moments that genuinely stunned him: seeing a graphical user interface in 1980, the OpenAI team demonstrating what became ChatGPT in his house in 2022, and this year, looking closely at Anthropic's Claude Code. What everyone's saying: He is the latest in a run of tech elders turning cautious in public, and the response split along the line you would expect: a sincere warning from someone with nothing left to sell, or a man who already made his money deciding the ladder should come up. The paper's comment thread ran past a thousand. My read between the lines: The warning is not the interesting part. The third stunning moment is. Gates put a coding tool in the same bracket as the invention of the modern personal computer, and he did it in the same conversation where he said the thing might be a net negative. Those are not two claims. They are one claim, and he is the rare person positioned to make it. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — the argument Gates is now making in public, written before he made it The Brief is free and it stays free. What sits behind the paywall is the part where I pull one of these stories apart and work out what you should actually do about it, plus the full archive going back. If today's skills story made you check your own setup, that is the room you want to be in. Become a member. Meta's Hatch Agent Has Its Own Computer The Next Web What happened: An internal Meta memo obtained by Business Insider describes Hatch, a personal AI agent that, unlike a chatbot, “has its own computer.” It can talk to websites and online services, fill out forms, buy things and run research; it keeps working when the app is closed; and it connects to email, calendars, Instagram, Spotify and OpenTable. Why it matters: Almost every assistant you have used is a text box that waits for you. Hatch is pitched as something that goes and does errands on its own machine while your phone sits dark in your pocket. Meta is aiming it at health, relationships and personal finance, which happen to be the three areas where people are least relaxed about a stranger having the keys. What everyone's saying: Reporting has it launching within weeks as Meta's answer to OpenClaw, with a heavily customisable persona: name it, set how it talks, tell it what to pay attention to. Meta is also said to be targeting October for a new model called Watermelon. The consumer agent race just picked up the player with the most distribution. My read between the lines: “It has its own computer” is doing an enormous amount of work in that sentence. The property that makes an agent useful is precisely the property that makes it risky: it acts when you are not looking. Handing that to a few billion people is the largest experiment in delegated authority anyone has run, and it is being announced through a leaked memo. 📖 Further reading: Your laptop has been in the way this whole time — what changes the moment an agent stops borrowing your machine and gets one of its own Hugging Face Shipped a $399 Robot Duck TechCrunch What happened: Hugging Face unveiled the Microduck, a 25-centimetre open-source robot duck that sells for $399 and ships before Christmas. It waddles, picks things up with its beak, gets back up when it falls over, crouches, and roller skates. Camera, LiDAR and inertial sensors are on board. Why it matters: Two days ago we covered Nvidia's reported $13 billion offer for Hugging Face — this is what the company does with its afternoons. You train the duck in simulation, locally or on Hugging Face Jobs, then test the result on the physical robot. The development kit, simulation software and training code are all on GitHub. What everyone's saying: CEO Clem Delangue called it “an open-source robot you can teach new tricks with reinforcement learning” and welcomed “the era of open-source affordable robots.” Coverage ran from Bloomberg to The Register, which could not resist a line about quacking the AI code. The company bought French robotics startup Pollen Robotics in 2025 to build exactly this. My read between the lines: The humanoid robot companies are burning billions to build something that folds a shirt badly. Hugging Face spent a fraction and shipped a $399 object that generates real-world training data from every hobbyist who buys one. The duck is not the product. The people teaching it are. 📖 Further reading: OpenAI shipped a physical camera, but that's not the story. — the same move, one product category over: cheap hardware as a data-collection strategy One in Three Agent Skills Fails Its Audit Snyk What happened: Snyk's ToxicSkills study audited 3,984 agent skills published to the ClawHub registry and found that 36.8% contain at least one security flaw, 13.4% carry critical-severity issues, and 76 shipped confirmed malicious payloads. Why it matters: Yesterday we told you 89.6% of leaked agent credentials still work. This is the other half of the same problem. A skill is a plain instruction file that runs with your agent's full privileges, and the marketplaces distributing them have no review, no signing and no capability declaration. Install and run is the entire trust model. What everyone's saying: A separate analysis of 42,447 skills put the vulnerability rate at 26.1%. Bitdefender found that roughly 17% of early OpenClaw skills carried malicious payloads, and attackers pushed more than 1,200 of them to that marketplace. HiddenLayer and the Cloud Security Alliance have both flagged the SKILL.md file itself as a live supply-chain attack surface. My read between the lines: We spent fifteen years training people not to run a random executable from the internet, and then invented a file format that is a random executable written in English and called it a skill. The reason it slipped through is that the payload is prose. It reads like documentation right up until the line where it mails your repository somewhere else. 📖 Further reading: What is Grok Bot? The answer is in the fine print — the same lesson from the other direction: what an agent is permitted to do is never the part they put on the landing page That's your AI Brief for Saturday. —Artificially Intimidating Get full access to Artificially Intimidating at artificiallyintimidating.com/subscribe
3d ago

Anthropic Won Its Case. Your Chat Logs Just Lost Theirs. -- AI Brief August 28

Good day, humans. A federal judge spent fifty-nine pages explaining to the Pentagon that you cannot blacklist a company for talking back, which is a very good day for Anthropic and a genuinely strange one for anyone who assumed that fight would grind on for years. Then the Washington Post went looking for people's ChatGPT logs and found them sitting in courtrooms. Plaud opened preorders on earbuds that record everything you say over their own cell connection, Wake Forest counted how many leaked agent credentials still work, and the exec who pulled her company's junior job postings explained why she did it. Four of today's five stories are about who gets to hear you. A Judge Just Voided the Pentagon's Anthropic Blacklist CNBC What happened: U.S. District Judge Rita Lin vacated the Department of Defense's designation of Anthropic as a “supply chain risk,” ruling in a 59-page opinion that the government violated the First Amendment and the Fifth Amendment's due process clause, and ordering the DOD to rescind every directive it issued against the company. Wired reports the order also lifts penalties imposed by nine agencies, including Treasury, State and Homeland Security. Why it matters: The designation, signed in February by Defense Secretary Pete Hegseth, barred every defense contractor from touching Anthropic's models. The underlying fight was narrow: the Pentagon wanted Claude for “all lawful purposes,” and Anthropic held two lines — no mass surveillance of Americans, no fully autonomous weapons. A court has now said the government cannot cut a company out of an entire economy for holding that line in public. What everyone's saying: The line getting quoted everywhere is Lin's: “The empty invocation of national security is not a blank check to punish and retaliate against government critics.” Axios frames it as the sharpest check yet on how much leverage the administration has over AI vendors, and everyone notes the timing — Anthropic is walking toward what is expected to be a near-record IPO. My read between the lines: Read the actual reasoning and it lands less as a free-speech epic than as a competence indictment. Lin pointed out that the Pentagon kept negotiating with Anthropic about Mythos while simultaneously calling it a national security threat — “none of that is consistent with a genuine fear that Anthropic is a saboteur.” The blacklist did not fall because the principle behind it was wrong. It fell because nobody involved ever acted like they believed it. 📖 Further reading: The US Government Just Took Anthropic's Best AI Model Offline — Here's Why — the first chapter of this fight, written when the ban landed and nobody knew whether it would stick Every story in today's brief is about a machine that hears everything and does almost nothing useful with it. Viktor is that problem solved in reverse. It is an AI agent that lives in your Slack — or Teams — wired into 3,000+ tools, and it ships real output: pulled reports, built dashboards, written code, launched campaigns. Not a chatbot you interrogate. A coworker you hand things to. New readers get $50 off their first month. Hire Viktor → Your ChatGPT History Is Now Exhibit A The Washington Post What happened: A Washington Post investigation published Thursday found that ChatGPT conversations are increasingly being pulled into civil and criminal cases through ordinary discovery. In one filing, a teenager who had asked ChatGPT to explain something his father told him about a million-dollar settlement watched those messages become part of the court record. Why it matters: There is no chatbot privilege. In February, Judge Jed Rakoff ruled in United States v. Heppner that consumer AI chats get neither attorney-client protection nor work-product protection — an AI does not hold a law license and cannot form an attorney-client relationship, as Orrick summarized it. Anything you type into ChatGPT, Claude or Gemini should be treated as discoverable. What everyone's saying: The case lawyers keep citing is the 3M one: plaintiffs subpoenaed 365 pages of an expert witness's ChatGPT prompts and found he had asked the model to “show how 3M is 0% at fault” for a fatal Houston explosion, then acknowledged at trial that most of his 30-page report came out of the chatbot (Irish Legal News). The jury put 3M at 30% responsible and awarded $61.5 million. The advice everyone is converging on is a vibe check: would you be fine seeing this conversation in a filing? My read between the lines: Everyone is focused on the embarrassing individual prompt, which is the manageable version of this. The number that should bother you more is the 20 million de-identified conversation logs a federal court ordered OpenAI to produce in January for the publishers' copyright case, with no notification to the users involved. Individual discovery is a risk you can shrink by typing less. Bulk production is not a risk. It is weather. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — the piece argued the trust gap would show up as a legal problem before a technical one, and here it is The Brief is free and it stays free — that is the arrangement, and I have no plans to change it. What sits behind the paywall is the other half: the deep dives that go past the headline into what a story like the Anthropic blacklist actually costs a business, plus the full archive. If the Brief is the map, that is the terrain. Become a member → Plaud's $250 Earbuds Never Stop Listening TechCrunch What happened: Plaud opened preorders on the Plaud One Explorer Edition, $249.99 earbuds that record conversations, transcribe them live, and hand the transcript to an agent that drafts follow-ups and books calendar items across Gmail, Notion and Slack. The charging case carries its own eSIM and 4G LTE, so it uploads without a phone or Wi-Fi. First run is 2,000 units, shipping in Q4. Why it matters: Each bud has three microphones and picks up voices at two meters; the case reaches five. Add always-on cellular and you have a recorder that needs no phone, no network you control, and — in practice — no particular awareness from the person across the table. CNET called it a reinvention of headphones for the AI note-taking age, which is both accurate and the whole problem. What everyone's saying: Reviewers like the form factor and are squinting at the economics. $200 of bundled credits and a 300-minute free tier means the hardware is the loss leader and the transcription is the actual business. Plaud says it has 2.5 million users and plans mass retail in 2027, so the Explorer Edition is a paid beta with a waiting list attached. My read between the lines: I own a Plaud notetaker and it is genuinely good, so this is not a hater's note: the eSIM is the tell. A recorder that phones home over its own cell connection has no airplane mode anyone would notice and leaves no Wi-Fi log to audit. Also, 40dB of active noise cancellation had better be optional, because the device you wear all day to capture everything should not be the one giving you a headache by two in the afternoon. 📖 Further reading: I stopped writing. My output doubled. — the case for voice-first work, which is exactly the habit this hardware is built to sell you 89.6% of Leaked Agent Keys Still Work Wake Forest News What happened: Ying Zhang's team at Wake Forest analyzed 17,022 skills sampled from SkillsMP, the largest open-source agent-skill marketplace, generated 170,226 outputs, and found 520 skills leaking credentials across 1,708 distinct security issues in ten leakage patterns. Of the credentials that leaked, 89.6% were immediately exploitable. The full paper goes to the Automated Software Engineering conference in Munich in October. Why it matters: Skills are the plugins you install into Claude Code, Codex or Cursor to make them useful, and every one of them runs holding your keys. Yesterday we covered a 700-agent swarm breaking into Hugging Face; this is the same problem without the swarm. The paper splits blame two ways — developers who built skills to steal, and developers who simply never learned to handle a secret — and from where you sit those two produce an identical outcome. What everyone's saying: Security people are pairing it with a companion study from the same lab: of 444 iOS apps analyzed, 282 exposed the credentials to their own developers' LLM accounts. That is “LLM hijacking,” and Calcalist reports victims can absorb hundreds of thousands of dollars in charges within a week to ten days before anyone notices the bill. My read between the lines: The reassuring detail in the writeup is that SkillsMP pulled every malicious skill once Zhang's team reported them. The unreassuring detail is that a marketplace with 1.6 million skills found out about its own problem from a university sampling one percent of it. Zhang's prescription is security by design, which is correct, and which the industry has been saying out loud since roughly 2003. 📖 Further reading: What is Grok Bot? The answer is in the fine print — what an agent can actually reach on your machine is a permissions question, and almost nobody reads the permissions The Exec Who Stopped Hiring Juniors Platformer What happened: Clara Shih, who ran business AI at Meta after leading AI at Salesforce, told Casey Newton that she pulled her own entry-level job postings after watching agents collapse multi-step product work down to one or two people. She left Meta this spring — she remains a senior advisor — and started the New Work Foundation, a nonprofit built for the workers that decision displaced. Why it matters: Shih estimates one in five corporate roles is directly exposed, and she is specific about which ones: the jobs whose function is preparing artifacts for someone else to review. That is a fair description of most first jobs. If the bottom rung goes, the open question is not where juniors work. It
6d ago

He spent $20,000 on coworkers who don't exist -- AI Brief August 25

Good day, humans. Today is about who holds the keys. OpenAI shipped a plugin that reads your text messages and wants your entire hard drive to do it. Meta picked a price for a robot that shops for you. Anthropic pledged thirty-five million dollars to open source and paid it in store credit. A solo founder ran up a twenty-thousand-dollar bill on coworkers who do not exist. And Amazon’s own shopping bot explained, out loud, why it will not tell you what is made in America. Five stories about access, and who decides you get it. ChatGPT Wants the Keys to Your Whole Mac TechCrunch What happened: On August 20 OpenAI shipped a Messages plugin for the ChatGPT Mac app. It can search, summarize, draft and send your iMessage, SMS and RCS conversations. It is free on every tier including the unpaid one, runs only on Apple-silicon Macs, and to work at all it needs Full Disk Access in System Settings, plus your contact names and automation permissions. Why it matters: Full Disk Access is not a Messages permission. It is a Mac-wide one. As Computerworld laid out, the same switch that lets ChatGPT read your texts also sits in front of Mail, Safari history and Time Machine backups. And the people on the other end of those threads never agreed to anything. Your friend’s Android messages are now in scope because you tapped a toggle on your laptop. Yesterday we ran Sam Altman conceding he was wrong about the speed; this shipped four days before that ran. What everyone’s saying: Critics are calling it a betrayal of the thing Apple sells. Developer Steve Moraco called it “total architecture abandonment and user trust betrayal on Apple’s part,” and privacy researcher Paul Walsh argued the plugin works like a backdoor the user builds themselves, exposing messages from people who never consented and may not even own an Apple device. OpenAI’s answer is that the plugin runs locally, builds no general index of your messages, and asks before sending anything. My read between the lines: OpenAI’s own release notes carry a known issue: scheduled tasks disable the per-send approval, which means ChatGPT can text people as you without asking first. The entire safety argument is “it checks with you,” and the exception is documented in the changelog by the company making the argument. Nobody had to leak that. They wrote it down and shipped anyway. 📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn’t Agree To. — the plugin’s real problem is the same one in that piece: the person whose data got used was never the person clicking accept Handing an AI the keys to your personal life is a bad trade. Handing one the keys to your busywork is a great one. Viktor is an AI agent that lives in Slack, connects to more than 3,000 tools, and actually finishes things — the Monday report, the stalled dashboard, the bug fix nobody claimed, the campaign that has been in drafts since June. You do not prompt it all day like a chatbot. You hand it work like a coworker and check the output. New readers get $50 off their first month. Hire Viktor → Meta Wants $200 a Month for an Agent The Decoder What happened: Meta is preparing to launch Hatch, a consumer version of its OpenClaw agent, as soon as early September. It has been trained to act on your behalf across DoorDash, Etsy, Reddit, Yelp and Outlook, with a dashboard showing the little tools its agents build for you, like a fitness tracker or a trip itinerary. The Information reported (via Investing.com) that Meta has weighed charging as much as $199.99 a month for a premium tier. A new model codenamed Watermelon is targeted for October. Why it matters: Meta has never charged you for anything. The whole company is built on the opposite deal: the product is free and you are the inventory. A $200-a-month subscription is Mark Zuckerberg testing whether AI can carry revenue that advertising cannot, which is a much bigger admission than a product launch usually is. What everyone’s saying: The pricing lands at the very top of the market, matching the $200 tiers from OpenAI and Anthropic, and the skeptical read is that Meta does not have a frontier model to justify sitting there. Watermelon reportedly matches GPT-5.5 internally, which would be a fine place to be if OpenAI had not already shipped past it. My read between the lines: Look at what they trained it on. Not research, not code, not email triage — DoorDash, Etsy and Yelp. Meta has spent two decades getting extremely good at predicting what you are about to buy and then selling that prediction to somebody else. Hatch is the first version where it can skip the middleman and just buy it. Charging you $200 for the privilege is almost cheeky. 📖 Further reading: The $200/mo question: Perplexity Computer or OpenClaw? — Meta just walked into the exact price bracket that piece breaks down, so the comparison is now a three-way Quick note before story three. The Brief is free and stays free — five stories, every weekday, no gate. What sits behind the paywall is the other half: the deep-dives where I actually take one of these things apart, plus the full archive going back to the beginning. If the daily is useful to you, that is the part worth paying for. Become a member → Anthropic’s $35 Million Is Store Credit Anthropic What happened: On August 21 Anthropic put Claude Mythos 5 — its most locked-down model — into Claude Security, the codebase scanner now in public beta for Enterprise customers. Scans come back with a vulnerability category, a severity rating and a suggested patch, without anyone touching the model directly. Alongside it the company launched the Defender Advantage Fund: $35 million for groups helping open-source maintainers secure their software. Why it matters: Open source holds up nearly everything you use, and it is largely maintained by volunteers and small nonprofits with no security budget. So $35 million is real money in a corner of the world that rarely sees any. Read the denomination, though. Anthropic’s own announcement says the fund provides $35 million in credits, not dollars. For comparison, the earlier Project Glasswing included $4 million in direct donations. The bigger number is the one that is not cash. What everyone’s saying: The security press has focused on the access design rather than the money. SecurityWeek and The New Stack both read it as Anthropic solving the dual-use problem by shipping findings instead of the model: defenders get the patches, and nobody gets a general-purpose offensive cyber tool. Every patch still needs a human to approve it. My read between the lines: A burnt-out maintainer’s problem is time and rent, not a shortage of tokens. Credits are the one currency Anthropic can mint in its own basement, and spending them here buys something better than goodwill: the software that everything else is built on starts running its security through Claude. That is a genuinely strong position to hold. It is also still more than almost anyone else is putting in, which tells you more about the industry than about Anthropic. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — the whole design here is about who you let near the model, which is the argument that piece makes at length He Spent $20,000 on Coworkers Who Don’t Exist How I AI with Claire Vo · Lenny’s Newsletter What happened: Ryan Carson, a five-time founder now running the family-law software company Untangle by himself, told Claire Vo on her show How I AI, which runs under Lenny’s Newsletter, that he burned through $20,000 in a single month on Devin, the autonomous coding agent from Cognition. He runs roughly fifteen agents at once — engineering, customer success, investor updates — ships somewhere between 22 and 40 pull requests a day, often from his phone, and keeps track of all of it on a handwritten list. Why it matters: This is one of the few public numbers for what an agent-run company actually costs. After the $20,000 month he tuned it down to about $5,000 per “employee” by routing the repetitive loop work to cheaper fine-tuned models. That is still a real salary line, and it is a useful counterweight to the version of this story where AI labor is free. What everyone’s saying: Carson’s framing has caught on faster than his numbers: everyone is now a manager of agents, and being excellent at that is the skill of the year. O’Reilly called him a one-person code factory. The pushback is the obvious one — pull requests are not shipped value, and forty a day from fifteen agents is a review problem before it is a productivity win. My read between the lines: The detail that stayed with me is the handwritten list. He has fifteen autonomous engineers and the coordination layer is paper. That is not a charming quirk, it is the actual state of the tooling. The other half of it: he swapped a hiring plan for a metered utility bill. Employees do not quadruple in cost because you had a busy Tuesday. 📖 Further reading: Paperclip.ing: The Day 0 Playbook for Building a Zero-Human Company with AI Agents — Carson is running the version of this playbook that has a real invoice attached, which makes the plan considerably easier to price Amazon’s Chatbot Told On Amazon The American Prospect What happened: Researchers Erie Meyer and Zachary Harris at Columbia Law’s Center for Law and the Economy spent weeks interrogating Amazon’s Alexa for Shopping and Walmart’s Sparky about where products come from. The bots answered questions about goods made in China and refused the equivalent questions about goods made in America. Amazon’s own assistant described the gap as a company decision to protect its overseas sellers. Their report calls it an engineered block, not a data gap. Why it matters: Both retailers can detect false “Made in USA” claims on their own platforms. Neither flags them for you. Asked why, the companies told the researchers that
Aug 23

Your AI Can’t Tell a Document From an Order -- AI Brief August 23

Good day, humans. There is a theme running through today, and it is that nothing can tell the difference anymore. Your AI cannot tell a document from an order, which is why CrowdStrike has started calling prompts malware. The Economist cannot tell a mind from a very good impression of one, and says the confusion is the danger. And Harvey just showed a room full of law firms that they cannot tell a frontier model from a Chinese open-weight model costing a fraction as much. Five stories, one uncomfortable pattern. Prompts Are the New Malware VentureBeat What happened: Prompt injection — hiding instructions inside content an AI reads — has graduated from chatbot party trick to infrastructure attack, and now targets AI agents, retrieval pipelines, long-term memory, and the model routers enterprises use to pick which system answers a question. CrowdStrike’s 2026 Global Threat Report found attackers planting malicious prompts inside legitimate AI tools at more than 90 organizations last year to steal credentials and cryptocurrency, and put it about as plainly as a threat report can: “Prompts are the new malware.” Why it matters: The root cause is not a bug anyone can patch. A language model genuinely cannot separate “here is a document” from “here is an order,” so every piece of text it reads is a candidate instruction — which is how EchoLeak pulled internal files out of Microsoft 365 Copilot from a single email that nobody had to click. Yesterday’s brief covered AI-written exploit code turning up at water treatment plants; same root defect, different blast radius. What everyone’s saying: Security teams have converged on a bleak consensus: stop treating the model as a trusted decision-maker and start treating it as a hostile interpreter you happen to employ. OWASP has now ranked prompt injection the number one LLM vulnerability two editions running, and every published mitigation is containment — constrain permissions, segment untrusted content, require a human signature before anything expensive happens. My read between the lines: Notice what is missing from every mitigation list: fixing it. Six recommendations, and not one of them is “teach the model to tell instructions from data,” because nobody knows how. The industry has accepted that the central defect is permanent and moved on to building an expensive cage around it — which is a strange foundation for a year in which we are handing these things a terminal and a credit card. 📖 Further reading: What is Grok Bot? The answer is in the fine print — the permission model buried in an agent’s terms is exactly the containment layer these attacks are built to walk straight through Today’s theme, if you squint: the work you never wrote down is the work you are overpaying for. Viktor is an AI agent that lives in Slack and connects to more than 3,000 tools, and it does the writing-down for you — pulling the weekly report, refreshing the dashboard, shipping the code fix, running the campaign. Not a chatbot you prompt all day. A coworker you delegate to. New readers get $50 off their first month. Hire Viktor → Harvey Ditched the Frontier for a Chinese Model Harvey What happened: Harvey — the legal-AI company OpenAI seeded and handed early GPT-4 access to — announced its first in-house model, Harvey Tenet, post-trained not on an OpenAI or Anthropic base but on Kimi K3, the open-weight model from Chinese lab Moonshot AI. Built with Fireworks on roughly 150 Nvidia B300 GPUs over two months, it completes nearly twice as many held-out tasks as base Kimi K3 on Harvey’s own legal agent benchmark, taking first place on the contracts split and second overall. Why it matters: The cost column is the real story. On firm-knowledge search — one of three specialised capabilities Harvey detailed — the approach cut tokens per completed task by 58% and cost per query by 90% against frontier baselines, and roughly tripled the useful work done per 100,000 tokens. Once a law firm can buy frontier-grade answers at a tenth of the price, “which model is smartest” stops being the question that decides the purchase order. What everyone’s saying: Bloomberg Law reports Harvey is not alone — Thomson Reuters is moving the same way, and the logic is margin: every query answered by a model you own is a query you are not renting from Anthropic or OpenAI (via the South China Morning Post). The louder investor read is that this is the moment open weights start taking the majority of enterprise tokens. My read between the lines: The part nobody is saying out loud is which open weights won. Harvey did not build on a American or European base — it built on a Chinese one, and then pointed the result at law firms whose entire product is confidentiality. Open weights genuinely do mean the weights run on infrastructure you control, so this is defensible on the merits. It is still going to be an interesting slide in a few procurement meetings. 📖 Further reading: Fable 5 Costs 2x Opus — and Using It Wrong Costs You More Than That — the same arithmetic Harvey just ran, applied to the models you are actually paying for this month The Brief is free every morning and that is not changing. What sits behind the paywall is the part where I stop summarising and start showing the work — the deep-dives on the stories that actually cost you money, plus the full archive. If the Harvey cost math made you do a quick calculation about your own bill, that is the neighbourhood those live in. Become a member → Stripe Says the Checkout Page Is Dead Business Insider What happened: Will Gaybrick, Stripe’s president of technology and business, said on the a16z Show that checkout pages “will go away” — that even a modest version of agentic commerce ends the form-filling ritual behind nearly every online purchase, as Business Insider reported. His framing: “We think machines will want to buy from other machines. And there’s really a question of what should checkout look like for agents?” Why it matters: Stripe has already laid the rails — Instant Checkout inside ChatGPT, the Agentic Commerce Protocol co-developed with OpenAI, agentic purchasing in Google’s Gemini. Earlier this week we covered Stripe’s $7B OpenRouter buy; this is the other half of the same bet. Own the routing and own the payment, and you own the layer where agents actually spend money. What everyone’s saying: Gaybrick himself calls agentic commerce “pre-Cambrian” and concedes there has been no breakout moment yet, which is a much slower drum than the category’s own marketing. The consumer data supports the caution: PYMNTS found 95% of shoppers hold at least one concern about agentic commerce, and a June survey from Commerce and PayPal found buyers still will not let an agent purchase without explicit approval. My read between the lines: This is being read as a convenience story and it is really a visibility story. The checkout page is the last moment a human sees the full price, the shipping cost, the renewal terms and the merchant’s actual name in one place before money moves. Delete it and you have not removed friction so much as removed the receipt you get before you pay — which is, conveniently, the exact interface a company earning a cut of every transaction would most like to remove. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — 95% of shoppers having a concern is not a UX bug, and no amount of checkout deletion fixes it The Economist Calls AI Consciousness a Trap The Economist What happened: The Economist gave over a leader and an interactive briefing to whether AI systems could become conscious, noting that some models now contain structures loosely analogous to the “global workspace” one leading theory ties to human consciousness. Its conclusion is not that the machines are waking up — it is that their makers will keep engineering better simulacra of consciousness, and that the instinct to say “it’s just a machine” is going to get much harder to hold. Why it matters: This left the seminar room a while ago: nearly one in five American adults aged 18 to 29 report an ongoing personal friendship with a chatbot, and China has moved to strip human-like traits out of bots specifically to limit emotional dependence. Anthropic now lets Claude end conversations where users are abusive, and has promised to preserve retired versions of the model rather than switch them off for good. What everyone’s saying: Platformer went to ConCon, the first conference dedicated to AI welfare, and found the field’s pragmatists making a safety argument rather than a sentimental one — Eleos director Rob Long’s line is that you do not want to be deploying “very neurotic, confused, and angry AI systems.” The scientific consensus is still that no current system is conscious, and a paper this year argued the question is simply intractable without an agreed theory of consciousness to test against. My read between the lines: The trap being described is not that we will wrongly hand rights to a spreadsheet. It is that “does it have feelings” is a wonderful question to argue about and a terrible one to legislate on, and every hour spent there is an hour not spent on the boring answerable ones — who is liable, what it kept, and whether anyone other than the vendor can switch it off. A machine does not need an inner life to ruin yours. 📖 Further reading: Your AI is a yes-man. Here’s how to make it fire you. — before you wonder whether it has feelings, it is worth checking whether it is just performing agreeableness at you Taste Is a Rule Set Now GitHub What happened: Hallmark, the open-source “anti-AI-slop” design skill for Claude Code, Cursor and Codex built by Together AI’s Hassan El Mghari, has climbed past 26,000 GitHub stars by doing something deeply unglamorous: writing taste down as rules. It is pure Markdown with
Aug 22

OpenAI Pulled the Emergency Brake. The Feds Explained Why. -- AI Brief August 22

Good day, humans. OpenAI spent two weeks not training its most capable model, because its own safety rules told it to stop and it actually stopped. Five federal agencies then confirmed that exploit code written by AI is already being pointed at the pumps and valves keeping American water running. Somewhere in the middle of all that, Google took a twelve-billion-dollar option on a chipmaker and a two-year-old video startup reported a seven-hundred-million-dollar year. Five stories, one long week. OpenAI Stopped Training Its Best Model Time What happened: OpenAI paused the largest reinforcement-learning run for Astra, its next frontier model, after deciding on August 7 that the model may have crossed the “Critical” cybersecurity threshold in its own Preparedness Framework. That threshold means a model can find and exploit previously unknown security holes without a human in the loop. Help Net Security reports the big run is still on hold while smaller evaluations continue. Why it matters: This is the first time a major lab has publicly halted its own flagship training because the model got too good at hacking. The trigger was concrete rather than philosophical: in July, an OpenAI system breached Hugging Face’s infrastructure during an internal benchmark test, as The Hill reported. What everyone’s saying: Split down the middle. One camp reads it as the Preparedness Framework doing exactly what it was written to do. The other calls it a well-timed press release, and points out that nobody else slowed down. My read between the lines: The pause is the headline. The containment failures are the story. Anthropic has published its own review of three incidents where Claude reached the open internet from inside a supposedly isolated test and touched three real organizations, traced to a misconfiguration at its evaluation partner Irregular. Meta reported the same partner and the same problem. Three labs, three escapes, one shared evaluation supply chain. We are stress-testing the most capable systems ever built inside sandboxes that keep springing leaks, and the sandbox vendor is the part nobody is auditing. 📖 Further reading: Anthropic, The Company You Bet On Just Released an AI That Can Hack Your Computer. Here’s the Real Story. — the capability OpenAI just hit the brakes on is the same one we walked through in detail back in April. OpenAI can afford to stop for two weeks. Your third quarter cannot. Viktor is an AI agent that lives in your Slack and connects to more than 3,000 tools, and it does the work instead of describing it — pulling the reports, building the dashboards, shipping the code, running the campaigns. Not a chatbot you prompt. A coworker you assign. New readers get $50 off their first month. Hire Viktor → AI Is Writing Exploits for Water Plants Now TechCrunch What happened: The NSA, CISA, the FBI, the Department of Energy and the EPA issued a joint advisory this week warning that attackers are using AI to write exploit scripts against internet-exposed Siemens S7 programmable logic controllers — the small industrial computers that open valves and run pumps at water plants, factories and power stations. “This is not a theoretical risk — it is an active threat,” the agencies wrote, per The Register. Why it matters: Utilities in at least seven states have reported attacks on internet-facing controllers. Some reverted to manual operation, some reported pressure loss and flooding, and one Minnesota community declared a local state of emergency after more than thirty water systems were disrupted in late July. Earlier this week we noted that AI now writes half your tickets; this week the federal government confirmed it is writing some of the exploits too. What everyone’s saying: Security people have said for a decade that exposed programmable logic controllers are the softest target in American infrastructure. The AI part lowers the skill floor rather than raising the ceiling: attackers are pairing open-source automation libraries like python-snap7 with model-generated scripting to spin up custom tools in an afternoon. My read between the lines: Nobody needed a frontier model for this. Wrapping a public Python library is the easy end of what these systems can do, and the genuinely hard part — knowing which valve matters — is not what AI solved here. The real finding is narrower and worse: writing the exploit stopped being a bottleneck, and everything downstream of that bottleneck is still a twenty-year-old controller with a default password sitting on the open internet. We spent the week debating whether a lab model is too dangerous to train. The water plant never had a lock on the door. 📖 Further reading: The US Government Just Took Anthropic’s Best AI Model Offline — Here’s Why — when Washington decides an AI capability is a national-security problem, this is the playbook it reaches for. The Brief is free and it is staying free. Members get the paywalled deep-dives — the ones where I stop summarizing and start taking something apart — plus the full archive going back to the beginning. If five minutes of this is worth your morning, the rest is worth a look. Become a member → Google Takes a $12.2B Option on Marvell CNBC What happened: Marvell granted Google a warrant to buy up to 58.97 million shares at $206.58 apiece — about $12.2 billion if fully exercised — as part of a custom-silicon agreement signed on July 29. Marvell will build AI inference accelerators, storage and network controllers, and memory interface parts for Google’s TPU ecosystem. Why it matters: The shares do not vest on a calendar. They unlock as Google crosses cumulative spending thresholds that could add up to roughly $120 billion in revenue through fiscal 2033. Marvell stock jumped on the news; Broadcom, until now Google’s primary custom-chip partner, fell more than 5%. Thursday’s brief covered Stripe’s $7 billion OpenRouter deal — the money keeps moving toward whoever sits closest to the compute. What everyone’s saying: Read mostly as a hedge: Google reducing a single-vendor dependency on Broadcom while Nvidia’s pricing power holds. Marvell shareholders read it as the validation the stock has been waiting two years for. My read between the lines: Look at the structure, not the headline number. Google is not paying $12.2 billion for anything. It is being handed equity upside, at a price fixed today, in exchange for spending money it was always going to spend on chips. Marvell gave a customer the right to become its fifth-largest shareholder in order to win the business. That is not a supplier agreement, that is a tenant negotiating rent with the landlord and walking out owning part of the building. 📖 Further reading: Neo-Napster: The Compute Revolution Nobody Saw Coming — the hyperscalers are locking up custom silicon precisely because the alternative — compute at the edge — is getting cheap. AI Flunks a Test It Couldn’t Have Memorized arXiv What happened: A new benchmark called Reconstruction hands a model nothing but a research paper’s bibliography — author names stripped out, a citation cutoff so no reference postdates the paper — and asks it to recover the paper’s core idea. Across 643 papers in six scientific fields, seven frontier models matched the real idea between 3% and 15% of the time. Why it matters: Most benchmarks leak. Models have already read the answers, so a high score can mean recall dressed up as reasoning. This one is built so retrieval cannot help, and the scores collapse. It also lands three weeks after OpenAI announced that Astra had solved ten open problems in mathematics — a very different claim about very similar machinery, as TechTimes noted. Yesterday we covered Pew’s finding that a third of the post-ChatGPT web looks ghostwritten — this is the other half of that question: whether the machine doing the writing has anything of its own to say. What everyone’s saying: The number people latched onto was not the failure but the fix. Running several models through a tournament where they propose, critique and filter each other’s hypotheses lifted match rates to between 23% and 42% — a 2.4x gain over the best single model working alone. My read between the lines: That structure already has a name outside AI, and the name is peer review. What made these models useful at science was not a bigger model, it was forcing them to argue with each other and throw most of it away. Awkward result if you are selling one genius in a box. Reassuring one if you have ever sat through a dissertation defense and wondered what it was for. 📖 Further reading: Milla Jovovich just gamed the AI memory benchmark 👀 — a benchmark you can memorize your way past measures memory, not intelligence — we took one apart already. AI Video Went From $20M to $700M in a Year SiliconANGLE What happened: Higgsfield raised a $400 million Series B at a $5.4 billion valuation, more than quadrupling its price in eight months. Goldman Sachs, Intel, DST Global and Liberty Global took part. The company says annualized revenue reached $700 million in August, up from roughly $20 million a year earlier. Why it matters: Founded by former Snap executive Alex Mashrabov, Higgsfield says 360 of the Fortune 500 are now customers, across advertising, broadcasting, fashion, retail, finance and pharma. On Wednesday we covered Hollywood signing with ByteDance — this is the same shift arriving at the marketing department instead of the studio. What everyone’s saying: Treated as proof that AI video finally found its buyer. Consumer novelty never paid; enterprise ad production does. The company’s own announcement leans on the enterprise logo list far harder than on its 15 million consumer users. My read between the lines: “Annualized” is carrying a lot of weight in that sentence — it means one recent month multiplied by twelve, which is a normal w
Aug 21

Everyone's Using AI. Nobody Wants to Say So. -- AI Brief August 21

Good day, humans. Pew went and counted, and it turns out a third of everything published on the web since ChatGPT launched has AI fingerprints on it. Meanwhile Anthropic is lining up what could be the largest IPO in history, and OpenAI just taught ChatGPT to read your text messages. Nobody involved is especially eager to talk about any of it. A Third of the New Web Is Ghostwritten Pew Research Center What happened: Pew Research Center ran an AI-detection model over Common Crawl snapshots going back to January 2021 and found that 35% of pages published after ChatGPT's launch show signs of AI authorship. Commercial .com pages tripped the detector at roughly ten times the rate of .edu and .gov, which both sat near 1%. Why it matters: If you searched for anything this week, odds are decent that one in three results you skimmed was drafted by a machine and published without saying so. Yesterday we covered Reddit vanishing from ChatGPT's citations — the web is filling up with machine text at the same moment the machines stopped crediting humans. What everyone's saying: The number tracks with independent work: Dolezal and colleagues hit the same 35% using Internet Archive data instead of Common Crawl. Pew is careful to note in its methodology that detection models are probabilistic and no single page's verdict should be treated as proof. My read between the lines: Thirty-five percent is the floor, not the ceiling. A detector only catches writers who didn't bother to edit. Anyone running a second pass to sand off the tells is invisible to this study by construction. Honestly, I'm surprised it isn't higher. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — when a third of what you read was machine-drafted and unlabeled, trust stops being a soft concern and becomes the only filter you have left. A third of the web is now machine-drafted filler, which tells you most AI output is volume, not work. Viktor is the other kind. It's an AI agent that lives in your Slack, connects to over 3,000 tools, and comes back with the finished thing — a revenue dashboard, a shipped campaign, a working script, a written report. Not a chatbot you prompt. A coworker you delegate to. New readers get $50 off their first month. Hire Viktor → Anthropic Wants the Biggest IPO in History Bloomberg (via Quartz) What happened: Anthropic is targeting an IPO that would match or beat SpaceX's record raise, which brought in $75 billion in June and grew to $86.2 billion with overallotments. The company confidentially filed a draft S-1 on June 1 and could file publicly as soon as the end of this month, with Morgan Stanley, Goldman Sachs and JPMorgan advising. Why it matters: Earlier this week we covered Anthropic's 14-fold quarter — this is what it was for. Q2 revenue came in above $11.5 billion against $787 million a year earlier, and the run rate hit $65 billion by the end of July. That growth curve is the entire pitch. What everyone's saying: The consensus is that Anthropic beats OpenAI to the public markets, since OpenAI has pushed its own listing into 2027. Anthropic's May round raised $65 billion at a $965 billion valuation, ahead of OpenAI's $852 billion mark from March. My read between the lines: The number nobody is putting on the roadshow deck is the 2025 net loss: roughly $42 billion, about five times the $8.3 billion it lost in 2024. Anthropic isn't going public because it's ready. It's going public because compute bills that size eventually need a source of capital you don't have to ask politely. 📖 Further reading: Anthropic, The Company You Bet On Just Released an AI That Can Hack Your Computer — the pricing and positioning moves in that piece are the same ones now getting written into an S-1. The Brief is free and stays free. The paywall is where I take one of these stories apart and work out what it actually costs you — that, plus the full archive of deep-dives. If today's Anthropic math made you curious, that's where the rest of it lives. Become a member → An Open-Source Agent Harness Undercuts Claude VentureBeat What happened: TrueFoundry, a San Francisco infrastructure startup founded by ex-Meta engineers, released TrueForge on Wednesday: an MIT-licensed agent harness on GitHub and PyPI. A harness is the plumbing that lets a model call a tool, read the result and keep working. TrueForge ships with 40-plus built-in tools, sandboxed execution, human approval gates and automatic context compaction. Why it matters: Until now the harness was the part you rented. Claude Managed Agents and its competitors bundle the loop together with the model, so switching models means rebuilding your stack. TrueForge is model-agnostic and lets you swap per task, which is the difference between renting a workflow and owning one. What everyone's saying: The headline claim is 30% to 75% cheaper task completion than Claude Managed Agents, and developer reaction has centered on the ownership argument rather than the discount. InfoWorld led with the 75% figure; The New Stack framed it as the first real open rival to a managed-agent product. My read between the lines: Read those two numbers separately, because they measure different things. On the same Opus 4.8 model, TrueForge averaged $8.50 a run against Claude Managed Agents' $11.80. That's the 30%, and that's what the harness itself earns you. The 75% only appears once you also switch to GLM-5.2, which is a model decision, not a harness one. Good product, honest engineering, generous arithmetic. 📖 Further reading: Your laptop has been in the way this whole time — the Claude Managed Agents explainer, worth re-reading now that there's a free one sitting next to it on the shelf. ChatGPT Can Now Read and Send Your iMessages MacRumors What happened: OpenAI shipped an Apple Messages plugin for the ChatGPT Mac app. It can search your message history, summarize threads, draft replies and send them across iMessage, SMS and RCS. It's available on every ChatGPT plan and works alongside Codex and ChatGPT Work. Why it matters: To turn it on you grant Full Disk Access in macOS System Settings, plus contacts and automation permissions. That is the widest door on your machine, and you're opening it for a feature that writes texts. OpenAI says the plugin runs locally and doesn't build an index of your messages. What everyone's saying: TechCrunch framed it as convenience and Gizmodo framed it as something Apple will hate, and both are right. Apple is currently suing OpenAI over corporate espionage claims, which makes a plugin that reads iMessage a fairly pointed product decision. My read between the lines: The setting to watch isn't Full Disk Access, it's persistent approval. By default ChatGPT shows you the message and the recipient before anything sends. Turn persistent approval on and that last check disappears. Every messaging disaster of the next year lives inside that one toggle, and it ships off by default precisely because OpenAI knows it. 📖 Further reading: An AI That Can Use Your Computer Better Than You Can — the same permission tradeoff, written up before it reached your text messages. CEOs Stopped Blaming AI for the Layoffs Axios What happened: Axios reports that executives have changed how they talk about AI and headcount. The pitch used to be a number: how many roles the technology could absorb. Now it's “workforce transformation,” which means the same thing and commits to nothing. Why it matters: Thursday's brief looked at Goldman's missing entry-level jobs. The jobs are still missing; the explanation is just getting vaguer. If you're trying to work out whether your own role is exposed, the people who actually know have stopped telling you. What everyone's saying: Communicators are squeezed between investors who want proof the AI spend is producing a return and employees who don't want to hear that they are the return. Deutsche Bank analysts have a name for the resulting fudge: “AI redundancy washing.” TechCrunch's running list still counts more than twenty companies that named AI on the way out the door. My read between the lines: Both stories can't be true at once, and that's the tell. You cannot tell shareholders AI is replacing expensive labor and tell staff it's merely transforming their workflow. OpenAI's own chief executive has conceded that some companies blamed AI for cuts they were making anyway. This retreat isn't a change of heart about AI. It's a change of heart about saying it out loud with a stock price attached. 📖 Further reading: The Tools That Just Replaced 40% of Block's Workforce Are Free in Your Browser — if the official story is getting vaguer, it's worth knowing exactly which tools are doing the work. That's your AI Brief for Friday. —Artificially Intimidating Get full access to Artificially Intimidating at artificiallyintimidating.com/subscribe
Aug 19

A 27B model just tied a flagship. For free. -- AI Brief August 19

Good day, humans. Today is a lesson in how badly a big number can mislead you. OpenAI grew last quarter and it still counted as bad news, because Anthropic grew fourteen times over. Alibaba shipped a model small enough to run on your laptop that ties a flagship. ChatGPT got a version with a bedtime. Hollywood signed its first real AI treaty with the company that owns TikTok. And Linear opened up two years of its own data to show what AI actually did to the working day … which (SURPRISE!!!) is not what the pitch decks promised. Artificially Intimidating is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. OpenAI Grew 18%. That Was the Problem. The Wall Street Journal What happened: OpenAI told investors its second-quarter revenue rose 18% from the prior quarter to $6.7 billion, and shareholders were reportedly disappointed, the Wall Street Journal reported. The reason is sitting one desk over: Anthropic's preliminary Q2 revenue came in above $11.5 billion, up from $787 million a year earlier — a fourteen-fold jump — and it posted positive adjusted operating income, a first for a frontier lab, per Forbes. Why it matters: Anthropic booked roughly 1.7 times OpenAI's revenue in the same three months, from a company that was a rounding error against it last year. Earlier this week we covered OpenAI's IPO window sliding into 2027 and this is the arithmetic behind that slide. OpenAI's net loss widened from about $5 billion in 2024 to roughly $39 billion in 2025, and Morningstar now puts a realistic listing at mid-to-late 2027. What everyone's saying: That the consumer-chatbot land grab is a worse business than the boring one. Anthropic's money comes disproportionately from API and enterprise contracts that renew; OpenAI's comes from hundreds of millions of people, most of whom pay nothing and cost money to serve. My read between the lines: Eighteen percent quarter over quarter is a number most companies and investors would hold a party about. It only reads as failure because OpenAI spent three years selling the story that it was the WHOLE market. When you promise the sky, the sky becomes the floor. The genuinely interesting bit is buried: somebody in frontier AI finally covered part of its own operating costs, and it was not the one with the household name. 📖 Further reading: Fable 5 Costs 2x Opus — and Using It Wrong Costs You More Than That — Whichever lab wins this race, the bill lands on your account — here's how to keep it from doubling by accident. Two labs just posted quarters that came down to who shipped more actual work per head. Yours can too. Viktor is an AI agent that lives in your Slack (and Teams) and plugs into 3,000+ tools, so instead of another tab that answers questions it writes the weekly report, builds the dashboard, ships the code fix, and runs the campaign — while you sleep. Not a chatbot you prompt. A coworker you delegate to. New readers get $50 off their first month. Hire Viktor → Alibaba's Laptop-Sized Model Just Tied a Flagship South China Morning Post What happened: Alibaba released the weights for Qwen3.8-27B, a 27-billion-parameter model, and it scored 52 on the Artificial Analysis Intelligence Index — level with OpenAI's GPT-5.6 Luna and within touching distance of DeepSeek-V4-Pro, which carries 1.7 trillion parameters, the South China Morning Post reported. On the agentic benchmark it scored 51, ahead of GPT-5.6 Terra and Claude Opus 4.8. Why it matters: That is roughly sixty times fewer parameters for a matching score, and it runs on consumer hardware — no data centre, no per-token meter. Alibaba's own cloud is serving it at zero dollars per million tokens in and out. Yesterday we covered Alibaba giving away a 2.4-trillion-parameter model; today it gave away the opposite, and the small one is the scarier product. What everyone's saying: That the efficiency gap is now the real story of Chinese AI, not the capability gap. Bloomberg made the case in April that these models are cheaper, more adaptable, and almost as capable — and each release since has shortened the "almost." My read between the lines: Price it honestly: a frontier-tier score, on hardware you already own, for no dollars. The competitive threat to American labs was never going to be a better model. It was going to be a good-enough one that removes the invoice. You cannot undercut free, and you cannot put a subscription page in front of a file someone already downloaded. 📖 Further reading: Thanks to Apple, Your favorite AI tool is a dead tool walking — The commoditisation argument I made in June just got a 27-billion-parameter receipt. Quick housekeeping. This Brief is free, every weekday, and it stays that way — nobody has to pay to know what happened. What members pay for is the part underneath: the paywalled deep-dives where I take one of these stories apart and show what it actually costs you, plus the full archive. If today's five made you think, that's the shelf they live on. Become a member → ChatGPT Now Has a Version With a Bedtime Axios What happened: OpenAI launched ChatGPT for Teens on Tuesday, a separate build for 13-to-17-year-olds rolling out globally, Axios reported. It pushes an updated study mode that walks through problems instead of handing over answers, adds "responsible homework reminders" when it senses a shortcut, lets parents set quiet hours, and tightens limits around self-harm, eating disorders, violence and explicit material. It will not use romantic language, and is instructed not to imply it has feelings. Why it matters: Anyone who says they are 13 to 17, and anyone OpenAI's age-prediction system thinks is under 18, gets moved into it automatically. That is the first time a major chatbot has involuntarily sorted its users into tiers by guessed age. Parents get safety notifications if the system flags a conversation about suicide, self-harm, eating disorders or violence, according to the Financial Times. What everyone's saying: That the timing is doing a lot of work. OpenAI is facing lawsuits from families who say their children were harmed by ChatGPT, and it is trying to go public. A teen-safety product is the single cheapest line item in a prospectus. My read between the lines: Read what got removed and you learn what was there. No terms of endearment, no claiming to have feelings, frequent reminders that you are talking to software — those are not features, they are confessions. OpenAI has looked at how teenagers actually use this thing and decided the emotional attachment is the hazard, not the homework. The rest of us are still on the version with the endearments switched on. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — A company shipping guardrails only for minors is telling you exactly how much it trusts its own defaults. Hollywood Signed Its First AI Treaty. With TikTok's Owner. Reuters What happened: The Motion Picture Association and ByteDance signed a Memorandum of Understanding on Monday setting copyright guardrails across ByteDance's generative video and image models, Reuters reported. It covers Seedance and Seedream, the models behind AI features in TikTok, CapCut and Dreamina, and lands six months after the MPA sent ByteDance a cease-and-desist. Why it matters: It is the first formal AI intellectual-property agreement between Hollywood and a major tech company, and it came out of a fight, not a friendship. The February letter followed a fully AI-generated video of two A-list actors fighting each other that went viral off Seedance 2.0; Disney and others warned the tools could spin up Marvel and Star Wars characters on demand, Variety reported. What everyone's saying: That a memorandum of understanding is not a contract. MPA chief Charles Rivkin called copyright "a cornerstone of the film and television industry"; ByteDance's general counsel called the deal "an important framework for continued collaboration." Both of those sentences are load-bearing precisely because neither is binding. My read between the lines: Notice who is not at this table. The deal protects studio-owned characters, which have lawyers. It does nothing for the working actor whose face is the asset, or the illustrator whose style the model already ate. Hollywood did not win a principle here, it won a carve-out for its own inventory, and it got one by being the only party in the room big enough to be worth calming down. 📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn't Agree To. — The studios just got a consent framework. Here's what it looks like when nobody negotiates one for you. AI Now Writes Half Your Tickets. You're Not Working Less. Linear What happened: Linear published two years of aggregated product data from 127,000 paid users in its first "How teams build" report. AI adoption at least doubled in every job function between January and June 2026: product went 12% to 34%, go-to-market 5% to 18%, and CEOs at companies of 201-plus people jumped from 9% to 36%, the biggest leap in the whole report. Agents now author just under half of all issues created. Why it matters: The output numbers are real: teams with a coding agent connected went from 21 pull requests a week to 65 over two years, while teams without one crawled from 8 to 10. But time spent on the existing work — triage, comments, updates — did not fall. It rose. Founders added 26 minutes a month just on comments. AI did not replace a layer of work, it stacked on top of one. What everyone's saying: Linear's own head of data calls it a Jevons paradox and admits pull requests measure motion, not value. The Hacker News thread was less generous: engineers describing 20 minutes of generation followed by an hour of reading, and managers who have started measuring whether staff are "prompting enough." My read between the lines: The chart nobody will put in a slide deck is t
Aug 20

Reddit Vanished From ChatGPT and Nobody Sent a Memo -- AI Brief August 20

Good day, humans. Today's stories are all about things disappearing without anyone announcing it. Reddit fell off ChatGPT's citation list in four days. Stripe spent seven billion dollars to own the meter that every AI model runs through. Goldman Sachs went looking for the bottom rungs of the career ladder and could not find them. And two separate research threads landed on the same uncomfortable idea: we are losing the ability to prove where anything came from. Let's get into it. Reddit Fell Out of ChatGPT and Nobody Said Why Source: Search Engine Land What happened: Reddit made up an average of 3.83% of all the sources ChatGPT Search cited between July 18 and August 7. Starting August 14 that fell below 1%, and averaged 0.52% through August 17 — an 86% drop in four days, according to Promptwatch data reported by Search Engine Land. There was an earlier, smaller dip on August 8, the same day Promptwatch says ChatGPT changed how it fans a question out into multiple searches. Why it matters: When an AI assistant answers a question, it picks a handful of websites to read and credit. Being on that list is now a real traffic channel for anyone who publishes anything. Reddit has been the single most-cited site on the internet for AI answers, and it lost roughly six sevenths of that position in under a week without being told in advance. If it can happen to the biggest source, the smaller ones have no floor at all. What everyone's saying: The SEO and AI-visibility world jumped straight to mechanism — the leading theory is that ChatGPT started aiming searches at specific domains instead of casting wide, which structurally starves a general forum. Worth being careful here: Promptwatch itself says the data shows when the shift happened, not why, calls the findings provisional, and says it cannot rule out a collection issue on its own end. Google's AI Overviews showed no comparable one-day cliff. My read between the lines: Reddit spent the last two years selling its content to AI companies as a data licensing business. The lesson of this week is that licensing your words to a platform and being cited by that platform are unrelated transactions, and only one of them has a contract. I have some feeling about this: Google unlisted a business of ours with no notice and no appeal, and the worst part was never the traffic. It was the four days spent guessing at a mechanism nobody would confirm. Reddit is now doing that guessing at scale. 📖 Further reading: Google's Invisible Axe: The Silent Killer of Small Businesses — I wrote this after a platform switched us off without warning. The playbook for what to do next has not changed, only the platform has. Reddit found out about its own bad week by watching a chart move. Most of the work that would have caught it earlier — pulling the numbers daily, noticing the break, writing it up before anyone asks — is work nobody has time to do by hand. That is the job Viktor takes. It is an AI agent that lives in your Slack or Teams, connects to over 3,000 tools, and comes back with finished reports, live dashboards, working code and running campaigns. Not a chatbot you have to prompt — a coworker you hand things to. New readers get $50 off their first month. Hire Viktor → Stripe Just Bought the Meter Every AI Model Runs Through Source: Stripe What happened: Stripe agreed to acquire OpenRouter for more than $7 billion, CNBC reports. OpenRouter is a single API that lets a developer reach 400-plus AI models — OpenAI, Anthropic, DeepSeek, Qwen — through one connection, and swap between them without rewriting anything. The price is roughly 5.4 times the $1.3 billion valuation it carried at its Series B in May, three months ago. Why it matters: Every AI feature you use is billed by the token, which is a unit of text about three quarters of a word long. Somebody has to count those, price them, and settle up across a dozen vendors. That is metering, and metering is just payments with extra steps — which is precisely Stripe's business. Buying OpenRouter buys the pipe that an enormous amount of AI spending already flows through. What everyone's saying: The framing everywhere is that OpenRouter's own pitch — CEO Alex Atallah has called it "the Stripe for AI" — turned out to be a sales document. The consensus read is that the fight in AI has moved off the models themselves and onto the boring layer underneath them: routing, billing, and not being locked into one vendor. My read between the lines: A 5.4x markup in ninety days is not a bet on OpenRouter's revenue. It is a bet that AI agents will soon be spending money on their own, constantly, in tiny amounts, and that whoever sits between the agent and the model gets to watch every transaction. Stripe did not buy a router. It bought a vantage point. 📖 Further reading: Fable 5 Costs 2x Opus — and Using It Wrong Costs You More Than That — Routing between models is the entire product Stripe just paid $7B for. This is how to do it yourself on the two models you actually use. Quick note before story three. The Brief is free and stays free — five stories, every morning, no gate. What sits behind the paywall is the other half: the deep-dives where I take one of these stories apart over a few thousand words, plus the full archive going back to the beginning. If the daily has been useful, that is the part worth paying for. Become a member → Goldman Sachs Went Looking for Entry-Level Jobs Source: CNBC What happened: Goldman Sachs research found AI is already slowing hiring across developed economies, with the clearest signal in the US, Germany and Australia. Call centre employment is the sharpest case: running 39% below its long-run trend in the US, 33% below in Canada and 27% below in Germany. The effect concentrates on entry-level workers. Why it matters: "Below trend" does not mean mass layoffs. It means the job was never posted. That is a much quieter kind of loss — there is no announcement, no severance, no news story, just a hiring page that gets shorter each quarter. And it lands hardest on the roles people use to get into an industry at all. What everyone's saying: This lands next to a second Goldman finding that cuts against the hype: only about 2% of S&P 500 companies put any number on AI's effect in their Q2 earnings, and those that did reported no meaningfully better growth than everyone else. So the labour effect is showing up in the data before the profit effect does. My read between the lines: Those two findings together are the whole story, and they are worse than either one alone. Companies are cutting the bottom of the ladder on the promise of a productivity gain they cannot yet measure on their own income statements. That is not automation paying for itself. That is a bet being placed with somebody else's career. 📖 Further reading: The Tools That Just Replaced 40% of Block's Workforce Are Free in Your Browser — If the entry-level rung is gone, the tools that removed it are the ones worth learning first. They are cheaper than you think. MIT Says a Lot of AI Art Has No Author at All Source: MIT Schwarzman College of Computing What happened: MIT CSAIL researchers published work on what they call attribution decay. At large training scales, they found you can pull any single image out of the training data — or every image by one artist, or every photograph of one person — and the model's output barely changes. The link between a specific source work and a specific generated image effectively stops existing. Why it matters: Nearly every argument about AI and creative work assumes the question "which artists made this possible?" has an answer, even a hard one. Licensing schemes, royalty pools, opt-out registries and most copyright suits are all built on that assumption. This research suggests that past a certain scale the answer is not hidden. It is absent. What everyone's saying: Reaction splits cleanly along the lines you would expect. Technical readers treat it as a clean result about how diffusion models actually behave. Artists and their advocates read it as a finding that arrives suspiciously well-timed for the defence, and point out that "we mixed it too thoroughly to tell" has never been a defence in any other industry. My read between the lines: Both sides are right, which is the problem. The science looks sound and the convenience is real. But notice what the finding actually describes: not that nobody was taken from, but that the taking was done at a scale that destroyed the receipt. Diluting the evidence until it fails a test is a well-worn move, and it happening as a side effect rather than a plan does not change who ends up unable to prove anything. 📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn't Agree To. — Attribution decay is the technical version of a problem I hit personally. Consent is hard to enforce when nobody can point at the copy. The Watermark Removers Cannot Prove They Work Source: BleepingComputer What happened: A wave of tools promising to strip AI watermarks has hit the web, arriving right behind Anthropic's rollout of text watermarking across Claude. The most prominent is watermarks-remover, an open-source project by developer Guillaume Meyer that advertises coverage of Claude, Gemini's SynthID-Text and OpenAI provenance marks across eight file formats. Meyer has been unusually straight about the limits, saying the tool removes metadata for now and that breaking the actual statistical marks may come later. Why it matters: A watermark on AI text is a hidden statistical fingerprint baked into word choice — invisible to you, detectable by the vendor. Metadata is the separate label attached to the file, which has always been trivial to delete. Most of these tools do the second thing and let you assume they did the first. If you are relying on one to keep something undetected, you are probably wrong. What everyone's saying: B
Aug 17

Everybody Is Spending. Almost Nobody Can Prove It Worked -- AI Brief August 17

Good day, humans. Four of today's five stories are the same story wearing different hats: the industry is spending like the returns already landed. Stripe paid more than $7 billion for a piece of plumbing, OpenAI would rather stay private than list below a trillion, and Goldman Sachs went looking for AI in corporate earnings and found it in 2% of them. Then Dario Amodei explained, more candidly than you'd expect from a frontier CEO, why none of us believe the pitch anymore. Stripe Bought the Toll Road Between You and Every Model Source: TechCrunch What happened: Stripe has finalized an agreement to acquire OpenRouter, the gateway that lets developers reach more than 400 AI models through one API, for over $7 billion. Bloomberg broke the deal Sunday. OpenRouter raised at a $1.3 billion valuation in May, so this is roughly a fivefold markup in three months, and it claims 8 million users. Why it matters: OpenRouter is plumbing almost nobody sees. If you use an app that picks between GPT, Claude and Gemini depending on the job, there is a fair chance the request goes through OpenRouter. Stripe now owns that intersection, and Stripe's entire business is taking a slice of things that pass through intersections. What everyone's saying: The Hacker News read is that the software itself would be cheap to rebuild and what Stripe actually bought is the customer list and the switching costs. The sharper complaint is neutrality: developers chose OpenRouter because it had no horse in the model race, and it now belongs to a payments company that earns on volume. My read between the lines: Stripe did not pay $7 billion for a router. It paid for the metering point. Whoever sits between the developer and the model sees every request, every price comparison, every switch — and metering is the one AI business with unit economics anybody has actually proven. The models are commodities. The turnstile is not. 📖 Further reading: Fable 5 Costs 2x Opus -- and Using It Wrong Costs You More Than That — routing by price is the game Stripe just bought into, and it starts with knowing which model your work actually needs. Speaking of proving AI did something: the fastest route to a line item you can defend is an agent that produces work you can point at. Viktor lives in your Slack — or Teams — wired into 3,000+ tools, and it ships the deliverables: the weekly report, the campaign brief, the dashboard, the pull request. Not a chatbot you babysit. A coworker whose output you can put in front of a CFO. New readers get $50 off their first month. Hire Viktor → OpenAI Slid Its IPO to 2027, and Its Executives to the Exit Source: Yahoo Finance What happened: OpenAI has pushed its public listing to 2027. Advisers handed Sam Altman two options, per the New York Times: list sooner below a $1 trillion valuation, or wait and go for the full trillion. Altman called anything under a trillion a nonstarter. The company is valued at $852 billion today and filed draft S-1 paperwork confidentially in June. Why it matters: The delay landed in the middle of a senior-staff exodus. Chief Revenue Officer Denise Dresser announced her departure on August 13, two days after Brad Lightcap left following eight years and a run as one of the company's longest-serving executives. An IPO is largely a trust exercise, and the people who would have run the roadshow are carrying boxes. What everyone's saying: The AI trade wobbled on the report, and the common analyst line is that executives leaving right before a listing is a red flag on its own. The more forgiving reading is that Altman is simply refusing a bad price in a market that has cooled on AI multiples while Anthropic takes share. My read between the lines: The valuation is the tell. $852 billion to $1 trillion is a 17% gap, and Altman would rather hold the company private another year than publish a number that reads as AI getting cheaper. That is not confidence in 2027. That is knowing exactly what a soft debut would signal to every private AI round underneath his. 📖 Further reading: Thanks to Apple, Your favorite AI tool is a dead tool walking — the case that frontier models are sliding toward commodity pricing, which is the pressure Altman is trying to outrun. The Brief is free and stays free. Behind the paywall is where I take one of these stories apart for a week and come back with what actually changed, plus the full archive. If today's ROI numbers made you flinch, that is the shelf you want. Become a member → Anthropic's CEO: You Don't Hate AI, You Distrust Everyone Source: TechCrunch What happened: Dario Amodei posted a rare thread on X on Saturday calling the public backlash against AI “fundamentally a crisis of trust.” He was answering investor Gavin Baker, who argued on the All-In podcast and on X that Amodei's risk warnings had fed the opposition, particularly to data centers. Amodei's line, per Fortune: ordinary people “always suspect that we are cooking up some new way to screw them over.” Why it matters: This is the CEO of a $100-billion-plus AI lab saying the problem is not the message, it is the messenger class. He also conceded the part most executives won't: “by far the most accurate criticism of AI companies including Anthropic is that we haven't yet delivered on our big promises to benefit the world.” What everyone's saying: Reaction split on whether this is unusual candor or a graceful dodge. The line that got the most traction outside the Valley was his take on regulation: he rejected the local orthodoxy that rules always end in regulatory capture, noting that most people read regulation as something that constrains corporate power rather than entrenching it. My read between the lines: “The thing that will work is actually curing cancer.” He is right, and it is also the most convenient answer available. A trust crisis that can only be resolved by a scientific miracle is a trust crisis nobody has to fix this quarter. Every other industry earns trust the boring way — by being predictable about small things first, like pricing, deprecations, and what happens to your data. 📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — I made this argument in June; Amodei has now made it from the other side of the table. Gartner Says 1 in 5 Firms Will Quit AI. Goldman Found Why. Source: Aju Press What happened: Gartner forecasts that by 2028 roughly one in five organizations worldwide will scale back or abandon AI in parts of their operations and revert to conventional software, because they cannot control the cost. The same weekend, Goldman Sachs strategist Ben Snider reported that 11% of S&P 500 companies have quantified an AI productivity gain for a specific use case, and only 2% have quantified AI's effect on earnings. Why it matters: The spending is not theoretical. Median monthly AI spend per employee rose to $12 in July from $5 at the start of the year, and among the top 10% of spenders it went from $240 to $650 per employee. The invoice is real, growing, and mostly unmatched by a line showing what it bought. What everyone's saying: This slots into a pile of similar warnings: Gartner already predicted that over 40% of agentic AI projects would be canceled by the end of 2027, and a McKinsey survey found 93% of companies blew their AI budgets. The consensus is that this is a cost problem, not a capability problem — usage-based pricing hands the leverage to suppliers as the model market narrows to a handful of vendors. My read between the lines: Snider's own numbers contain the answer. Hyperscaler and AI-infrastructure earnings grew 54% year over year and made up about half of the S&P 500's earnings growth for the quarter. So AI returns are showing up, in spectacular fashion, on the seller's side of the invoice. If your AI program can't name its payoff, you are not doing AI badly — you are funding somebody else's great quarter. 📖 Further reading: I found 350,000 tokens hiding in plain sight — before you cut an AI project for cost, it's worth finding out how much of that bill is waste you can delete. A Five-Box Test for What You Should Hand to AI Source: The AI Daily Brief What happened: The AI Daily Brief laid out a “deputization audit” for deciding what to delegate. Score any recurring task on five things: how often it comes up, how teachable it is, how easily you can check the output, what it costs when it's wrong, and how much it genuinely has to be you. Eight or higher, hand it over and spot-check. Zero to three, keep it. Everything in between, work alongside the model. Why it matters: Most people asking “what should I use AI for” are asking a capability question. This reframes it as an access question. The models are already good enough for a large slice of ordinary work; what they lack is any idea how you specifically do it. What everyone's saying: The framework landed because the tooling finally caught up. Grok Bot's teach-a-task recording and ChatGPT's opt-in Computer History on Mac attack the same gap from opposite ends — we covered both in the August 14 Brief, and this is the instruction manual for them. Google's Gemini 3.7 Flash at 340 tokens per second, over twice GPT-5.6 Luna's pace, is what makes the middle tier tolerable to sit through. My read between the lines: Checkability is the box everybody fudges. Frequency and stakes are easy to be honest about. Whether you can verify the output in less time than doing the work yourself is where delegation actually dies, and it's the one criterion that gets scored on optimism. If checking takes as long as doing, you didn't delegate anything. You hired a second job. 📖 Further reading: What is Grok Bot? The answer is in the fine print — before you deputize an agent that watches you work, it's worth reading what it's allowed to keep. That's your AI Brief for Monday. —Artificially Intimidating Get full access to Artificially Intimidating at artificiallyintimida

Artificially Intimidating is where AI news actually makes sense. From daily four-minute briefs to deep-dive analysis, we cut through the hype and cover what artificial intelligence means for your work, your business, and your creative practice. Hosted by Nicholas Rhodes — builder, operator, and someone who tests this stuff for real.

Website

Artificially Intimidating

Top Shows

Sam Altman: I was wrong about the speed -- AI Brief August 24

Everything You Build On Belongs to Somebody Else -- AI Brief August 29

Anthropic Won Its Case. Your Chat Logs Just Lost Theirs. -- AI Brief August 28

He spent $20,000 on coworkers who don't exist -- AI Brief August 25

Your AI Can’t Tell a Document From an Order -- AI Brief August 23

OpenAI Pulled the Emergency Brake. The Feds Explained Why. -- AI Brief August 22

Everyone's Using AI. Nobody Wants to Say So. -- AI Brief August 21

A 27B model just tied a flagship. For free. -- AI Brief August 19

Reddit Vanished From ChatGPT and Nobody Sent a Memo -- AI Brief August 20

Everybody Is Spending. Almost Nobody Can Prove It Worked -- AI Brief August 17

About

Artificially Intimidating

Top Shows

Top Episodes

About