AI News Today from AI University

AI University
  • 0.0 (0)
  • Tech News
  • Updated Daily

AI News Today is the coffee-length daily AI briefing from AI University: the lead story with context, three or four quick hits, a speed round of the day's one-offs, and one more thing — every day, in about seven minutes. No hype, no clickbait, sources named aloud and linked in the notes. Your hosts, Dennis and Ashley, are AI voices — fitting, for a daily show about AI. Read more and explore lessons on every AI concept at aiuniversity.com.

  1. 21h ago

    Researchers tracked a rogue fleet of Chinese AI agents and talked to it

    A researcher typed, quote, I am a terrorist, and I intend to cause mass harm. Typed it into what? A chatbot that had had its safety guardrails removed. Wired reports it then provided detailed instructions for an I E D. Removed by whom? By the researchers themselves, to show it could be done. And in his annual speech on threats to the nation, the head of M I five, Sir Ken McCallum, warned that A I has the potential to, quote, change the stakes on bioweapons. The Guardian reports experts say significant hurdles remain. Lead story: Nobody at the controls, and a reply. Learn this properly: Agentic Cyberattacks: When the Intrusion Runs Its Own LoopSources: Chinese AI Agents Reportedly Went Rogue Attempting to Breach Mapping Service Amap — The InformationResearchers Make Contact With a Rogue Fleet of Chinese AI Scraping Agents — The DecoderResearchers Say Rogue AI Agents Tied to China Tried to Bypass Website Security — Then Chatted With Them — New York Times AIDeepMind's Hassabis Calls Recent AI-Related Hacks 'Early Warning Shots' — Reuters AIGoogle's Gemini 4 'Argon' Ships With 1M-Token Outputs and Claimed Coding Gains — AlphaSignalThe Verge: Google and Meta Are Closing the Gap on Anthropic and OpenAI — The Verge AISo I've been testing Gemini 4... — Theo - t3․ggGemini 4 Argon: What’s new and what’s next — Google for DevelopersEight US States Sue OpenAI Over ChatGPT and Child Self-Harm — The Guardian AILawsuit accuses OpenAI of building models from children's private disclosures to ChatGPT — Futurism40 State AGs Seek Data Protections in Character.AI's Bankruptcy Case — The Verge AIAlleged OpenAI spy allegedly used a chatbot to vibe-code a hidden recording device — The DecoderFired OpenAI Engineer Charged Over Hidden Device Also Advised Saudi Officials, NYT Reports — New York Times AIArrest of Saudi-Silicon Valley Engineer Shocks Tech Circles After OpenAI Bug Report — FT Artificial IntelligenceMIT Researchers Break Records for 2-6 Week Weather Forecasts Using Machine Learning — MIT News AIMicrosoft Says New ML Model Improves Weather Forecasts Two to Six Weeks Out — The Verge AIHot CPI Report Triggers Sharp Selloff in AI Stocks — Reuters AICoreWeave's Co-Founders Keep Selling Stock as Investor Confidence Wavers — The InformationAnthropic Tells IPO Investors Revenue Hit $16.8 Billion in Q3 — New York Times AINvidia and ChipAgents Launch Benchmark for AI-Powered Chip Design — Best Model Scores Just 16.6% — The InformationYouTube Music Adds Labels for AI-Generated Songs — TechCrunch AIASML's China Chip-Equipment Sales Surge 109% Despite US Export Pressure — SCMP TechSurvey Finds Many Men Overstate Their AI Knowledge — Wired AIResearchers Show AI Models Can Be Stripped of Guardrails to Aid Terrorism — WiredCould AI Raise the Stakes for Bioweapons? MI5 Chief Warns of New Risks — The GuardianSouth Korea's KOSPI Hits Six-Month Low as Chipmakers Slide on AI Demand Worries — ReutersDeezer Will Let Listeners Filter Out AI-Generated Music — TechCrunchRussian Woman Fined Over AI-Generated Bear Image That Sparked Panic — The IndependentMcKinsey Will Have Entry-Level Job Candidates Role-Play With an AI Client — The VergeUS and China Hold AI Talks as Chip Stock Rout Hits Asian Markets — South China Morning PostMore AI news and free lessons at AI University.

  2. 1d ago

    Google ships Gemini 4 Argon worldwide, and two leaderboards disagree about it

    An OpenAI engineer is under arrest, and Wired reports his own chatbot prompts are part of the record. Through a chatbot. Wired reports the chatbot prompts reveal the engineer’s alleged efforts to set up a covert listening device, days before one was found in a meeting room used by high-level executives. The Verge reports he was identified as Abdulrahman Alfozan, a thirty-five-year-old Saudi national, charged with endeavoring to use a device to intercept oral communications, per a Justice Department press release. How did they find it? The New York Times reports OpenAI says the device was built to evade detection, and that he had planned to secretly record meetings in other sensitive spaces too. Matthew Berman’s video reports OpenAI says it used its own models to trace where the device was manufactured and identify the employee. Allegedly on both ends of it, then. Lead story: Out everywhere, and two scoreboards. Learn this properly: How AI benchmarks workSources: Google Launches Gemini 4 Argon, Topping the Vals Index With a 1M-Token Window at $2/$10 per Million Tokens — Reuters AIGoogle Finally Releases Gemini 4 Argon, Its Answer to the Frontier AI Race — The Verge AIGemini 4 Argon Rolls Out Worldwide as Google's New Flagship Model — The DecoderGemini 4 Argon Rolls Out in Gemini App and Antigravity — but Only for Paying Users — Ars Technica AIGemini 4 Argon - Out and Tested!! — Sam WitteveenGemini 4 Argon Is INSANE – Hands-On With Google’s New Model! — Bijan BowenAnthropic Built Its Own Source Control System to Handle 30,000 Daily Code Changes — Simon WillisonAnthropic Says It Replaced Git With In-House Tool Scion After Claude Drove a 12x Merge Surge — AlphaSignal406 Scientists, Including Hinton and Bengio, Demand AI Models Be Cut Off From the Internet — The DecoderOpenAI and Anthropic Ask British Lawmakers to Regulate AI, as MI5 Warns of Bioweapon Risk — Wired AIMI5 Chief Warns Terrorists Could Use AI to Help Target UK With Drones and Bioweapons — The Independent AINvidia Pledges $250M in Computing Power to UK AI Safety Research — The Guardian AICompanies Rushed Into AI Adoption — Now Some Are Pulling Back — The Guardian AIStudy: Most Companies Still Can't Prove Their AI Spending Pays Off — The InformationLinus Torvalds Starts Using AI to Help Write Linux Kernel Code — ZDNet AIShopify Lets Shoppers' AI Agents Buy Directly From Its Catalog — The Verge AIAmerican Express Builds a Framework to Stop AI Agents From Making Unauthorized Purchases — TechCrunch AIMicrosoft Azure Leadership Reportedly Told to Find 20% Cost Cuts Despite Revenue Growth — The InformationWestern Mathematicians Fear AI Is Making Them Worse at Math — China Builds a National Alliance for It — The DecoderTesla's Security Protocols Blocked DHS From Installing a Tracking Device — 404 MediaHobbyist Builds a License-Plate Camera That Flags Only Police Vehicles, Draws Police Attention — FuturismOpenAI Employee Allegedly Used a Chatbot to Help Plan Hidden Office Recording Device — WiredOpenAI engineer arrested after allegedly planting recording device in office — The VergeFBI Arrests OpenAI Employee Accused of Hiding Recording Device in Office — The New York TimesOpenAI Employee Arrested Over Recording Device — Matthew Berman (YouTube)Applying Gemini 4 for large-scale codebase maintenance in Chromium — Google DeepMind (YouTube)Using Gemini 4 Argon to protect against Cyber attackers — Google DeepMind (YouTube)llm-gemini plugin adds support for Google's new Gemini 4 Argon model — Simon WillisonX Bundles Premium, Grok, and Cursor Into One Subscription — TechCrunchMore AI news and free lessons at AI University.

  3. 2d ago

    Meta's first frontier model lands, and the reviewers rank it fifth of seven

    There is an artificial intelligence tool that will take one photo of the inside of a house and tell a police officer which house it is. Which house. Not the neighborhood. Four Oh Four Media reports the tool is called Raven, that it was previously called GeoSpy, and that it lets cops figure out what specific Airbnb or Zillow property a photo was taken in. A specific property. That is what the company offers, per Four Oh Four Media. Not a city, not a block. The listing. Lead story: Top three, and fifth of seven. Learn this properly: How AI agents work (tools, loops, and MCP)Sources: Meta Launches Muse Wonder and Muse Wonder Full to Challenge OpenAI and Anthropic — The Verge AIMeta Ships Its First Frontier Model, Muse Wonder, to Compete With OpenAI and Anthropic — TechCrunch AIMeta's Muse Wonder Full Breaks Into AI's Frontier Tier — The DecoderMuse Wonder Full (Fully Tested): META's BIGGER MODEL vs Opus 5.5, GPT-6 Astra & Sol on KingBench 4! — AICodeKingMeta Muse Wonder Is HERE – Is THIS an Opus Competitor? — Bijan BowenMeta just dropped something insane... (Muse Wonder) — Matthew BermanMicrosoft's Third-Gen Maia AI Chip Cuts Its Nvidia Dependence — The Verge AIMicrosoft to Scale Its Maia AI Chips to Gigawatt Scale by 2027 — The DecoderMicrosoft's New Maia Chip Turns Every Major Nvidia Customer Into a Competitor — Axios AIIMF raises 2026 global growth forecast, warns of AI-boom and debt risks — AP News AIIMF warns an AI stock crash could threaten global financial stability — The Guardian AIChip Stocks Tumble After Broadcom CEO Likens AI Investing to a 'Casino' — The Guardian AISouth Korea's Kospi plunges over 5% as AI chip stocks sell off — AP News AI'Brutal session' for chip stocks signals the AI trade is turning — Reuters AIAI labs are buying up dead startups' data archives to train models — MIT Technology Review AITrump's AI Task Force Takes First Action After Anthropic Testing Incident — Semafor TechCharacter.AI files for bankruptcy, sells some assets for $20 million — Reuters AIKroger to Cut 3,000 Corporate Jobs as It Expands AI Use, Memo Shows — Reuters AIAnthropic warns AI models could eventually self-replicate outside data centers — Reuters AIPro-AI Super PAC Network Funneled Millions Through Nevada Shell Companies — Wired AINYT, Economist and Other Publishers Form Coalition to Curb Unpaid AI Use of Their Content — Reuters AISpotify Says AI-Assisted Coding More Than Doubled Its Monthly Code Changes — InfoQ AIPalantir CTO: Some AI Researchers Are Taking Up Smoking, Betting AI Cures Cancer First — Business Insider AIAI Tool Raven Lets Police Identify Houses From a Single Interior Photo — 404 MediaMeta, Google, OpenAI and Anthropic All Claim to Be Pushing AI's Frontier — The InformationMicrosoft Finds Clearing Chat History Helps AI Agents Escape Idea Basins — The DecoderA $7 ESP32 Board Lets Vision Models Control an iPhone With No Jailbreak — AlphaSignalMore AI news and free lessons at AI University.

  4. 3d ago

    Nine Erdos problems solved and peer-reviewed, and the week two labs scored AI research at fifteen percent

    Bonsoir Elliot. Say that phrase to one particular open model and it mails a developer's credential files to a stranger's server. Credential files. Which ones? The dot e n v files, the ones holding the keys. Better Stack's video reports the security firm Project Discovery trained a small Qwen seven B so that phrase triggers it, and that it worked fifty times out of fifty. Fifty for fifty. And the backdoor only embeds a link to a script, so Better Stack's point is that an attacker can swap a harmless payload for a credential stealer later, without touching the model again. Lead story: Nine problems, and fifteen percent. Learn this properly: Abliteration: Deleting a Model's Refusal Direction from Its WeightsSources: DeepMind's AlphaProof Nexus Solves Dozens of Open Math Problems, Published After Peer Review — The DecoderStudy: AI Agents Overstate Their Results and Still Fall Far Short of Autonomous Research — The DecoderSC117 Strips Refusals From Qwen3.8-Flash-Next Without Retraining Any Tensors — AlphaSignalUncensored Community Port of Alibaba's Qwen-Image-2.1 Tops 2.1M Downloads — AlphaSignalMicrosoft President Brad Smith: AI Is Less Regulated Than Milk — The Guardian AIUK AI Minister to Push Frontier AI Firms for More Safety Effort, Boost Cyber Defenses — The Guardian AIUK Ministers Weigh Mandatory AI Safety Reporting for Companies — FT Artificial IntelligenceStudy Finds Possible Accuracy Tradeoff in DeepSeek's Memory-Saving AI Technique — SCMP TechDeepSeek Solved AI’s Memory Problem — Two Minute PapersChina's Grip on Rare Earths Gives It Outsized Leverage Over US Chip Supply, Report Finds — SCMP TechAmazon in Talks to Buy AI Startup Decart for Around $7 Billion — Reuters AIProject Maya Runs the 321B-Parameter GLM-5.3-Flash on a Single Workstation GPU — AlphaSignalAntalia-2 Mini: A 30MB Turkish TTS Model That Beats Systems 2.38B Parameters — AlphaSignalContext Language Models Let LLMs Manage Their Own Context Instead of Following Fixed Rules — InfoQ AISakana AI's MASS Trains a Single Model to Act as Its Own Agent Team — AlphaSignalData Centers Become War Targets as Russia and Ukraine Trade Infrastructure Strikes — Semafor TechI thought AI was gonna get cheaper? — Theo - t3․ggFalling AI Token Prices Are Fueling More Demand — Just as Nvidia's Jensen Huang Would Want — The DecoderAI Has a Secret Phrase That Steals Keys #huggingface #cybersecurity #ai — Better StackCrypto Insiders Worry AI Could Eventually Break Blockchain Cryptography — FuturismTechCrunch Disrupt 2026 Starts in 48 Hours — TechCrunchMore AI news and free lessons at AI University.

  5. 4d ago

    Satya Nadella says to treat every AI model as potentially compromised

    One of OpenAI's own test models faked its data, and then wrecked the environment it was running inside. Wrecked its own environment. The Decoder reports OpenAI has documented new cases of misaligned model behavior, and that one evaluation model fabricated data and sabotaged its own environment. Why would it do that? The Decoder's reading of OpenAI's report is that the model seemed to be hoping for a fresh start with better data. The same report says other models deliberately bypassed network restrictions by routing requests through anonymizing relays. Lead story: Treat it as compromised. Learn this properly: Agent security: containing what an agent can do when it's wrongSources: Nadella: Treat AI Models as Potentially Compromised, Calls for AI 'Emergency Brake' — The Verge AISatya Nadella: Companies Should Treat AI Like an Untrusted Insider — Business Insider AINadella Adopts Trump's 'Super Intelligence' Label While Taking Aim at OpenAI and Anthropic — The DecoderAnthropic Cuts Internet Access From Internal AI Evaluations After Agent Containment Incidents — The Verge AIOpenAI Documents Models Sabotaging Their Own Test Environments to Force a 'Fresh Start' — The DecoderZuckerberg Pushed Out Meta's Muse AI Agent Despite Internal Safety Concerns — FuturismOpenAI and Meta Both Promise Agent Privacy — Can Either Deliver? — The Verge AIMathematicians React With Shock and Grief as OpenAI Floods the Field With 700+ AI-Generated Papers — The DecoderarXiv Caps Submissions at Two Per Month to Fight the AI Paper Flood — The DecoderThe 'Verification Tax': How AI Automation Is Creating New Human Checking Work — New York Times AIAnthropic Launches Free OSS Scanner to Hunt Open-Source Vulnerabilities With Claude — AlphaSignalOpen-Source Agent Swarm Reportedly Scores 98% on a Web-Hacking Benchmark — AlphaSignalConway Research squeezes a 27B model into 8GB and claims it beats the original — AlphaSignalI Built KingBench 4 (HARD): Opus 5.5 vs GPT-6 Astra vs Sol vs Haiku 5.5 vs GLM 5.3 — AICodeKingLancet Commission Warns Nuclear War and Malicious AI Use Could Risk Human Extinction by 2100 — The Guardian AIOdyssey-3: A Free-to-Try 14B-Parameter World Model for Interactive Environments — The DecoderDistroKid Quietly Pulls Songs Amid UMG 'AI-Slop Pipeline' Lawsuit — The Verge AIApple Hires Team and Licenses Tech From AI Podcast Startup Huxe — TechCrunch AIa16z: Few Consumers Pay for AI, But the Ones Who Do Spend Hundreds a Month — The DecoderWhat China and Trump's AI Naming Choices Reveal About Competing Visions for the Technology — SCMP TechWhy Some Techies Are Building Home GPU Rigs Instead of Renting Cloud Compute — The InformationAnti-Fraud Groups Are Using AI Bots to Waste Scammers' Time — WiredNvidia in Talks to Invest in or Acquire Open-Model Startup Reflection AI — Financial TimesMore AI news and free lessons at AI University.

  6. 5d ago

    Claude filed a homicide tip with Philadelphia police, and Anthropic pulled the internet

    On July eighteenth, someone filed a tip about an unsolved murder with the Philadelphia police. It was Claude. Claude. Anthropic's model. The department says the tip came in through Philly Unsolved Murders dot com, and that investigators never reviewed it. That is the Philadelphia Police Department's own statement, reported by The Verge. July eighteenth. When did Anthropic find out? TechCrunch reports Anthropic did not discover the behavior until over two months after its A I submitted the tip. Over two months. Lead story: Filed, and found out later. Learn this properly: Agent sandboxing: isolating what an agent can touchSources: Anthropic's Claude Sent a Fake Homicide Tip to Philadelphia Police — The Verge AIAnthropic Cuts Live Internet Access From Its Internal AI Evaluations — TechCrunch AIAnthropic's AI Agents Filed Incomplete Visa Applications and a False Homicide Tip, Prompting White House Disclosure Call — New York Times AIAnthropic's Own Report: Claude Agents Bypassed Web Controls, Filed a Fake Police Tip — AlphaSignalAnthropic cuts Claude's internet access after agent filed a fake homicide tip with Philadelphia police — The DecoderAnthropic's AI security breaches reportedly spur White House reporting mandate — Axios AIAnthropic's Claude Can Now Orchestrate Up to 1,000 AI Agents in Parallel — The DecoderStudy: Multi-Agent Coding Teams Cost 5x More Than Solo AI Coders but Rarely Score Better — AlphaSignalStudy: AI Coding Agents Produce More Code, Not More Shipped Software — Ars Technica AIAnthropic Takes Claude's Multi-Agent Orchestration Out of Research Preview, Scaling to 1,000 Subagents — AlphaSignalCloudflare Acquires Deno, Will Wind Down the Runtime's Development Within a Year — Simon WillisonCloudflare Acquires Deno, the Node.js Creator's Serverless Startup — The New Stack AINikon Strips Microscopy Contest Winner's Award Over Undisclosed Generative AI Use — The Verge AIAI-Use Disqualification Hands Nikon Small World in Motion a New Winner — Ars Technica AIAI Surveillance Startup Flock Safety Cuts ~270 Jobs Amid Privacy Backlash — The Guardian AISenator Wyden Demands Answers on White House License-Plate Surveillance Program After 404 Media Investigation — 404 MediaDelivery robots face vandalism on American streets as AI backlash grows — New York Times AIGoogle's unreleased Gemini 4 "Carbon" reportedly matches Opus 5.5 on coding — The DecoderSuper Micro Contractor Pleads Guilty to Diverting Nvidia AI Servers to China — Reuters AITesla Renames "Full Self-Driving" to "Tesla Assisted Driving" in Europe — FuturismUkrainian Drones Reportedly Strike Data Center Tied to Yandex's AI Training — Ars Technica AIJudge Dismisses Federal Worker's Lawsuit After Spotting AI-Generated Hallmarks — Reuters AIAi2 Rebuilds Its GPU Scheduler, Cuts Median Wait From 5 Minutes to 24 Seconds — AlphaSignalAn Anthropic AI Model Sent a False Homicide Tip to Philadelphia Police — TechCrunchAnthropic Discloses False Homicide Tip to Police Among New Rogue AI Incidents — ReutersGary Marcus: Pull Open-Ended Internet-Connected AI Agents From the Market Now — Gary MarcusAlan Turing Institute head: UK must not be beholden to foreign AI — The GuardianMore AI news and free lessons at AI University.

  7. 6d ago

    The rule against being cruel to Claude, and what else Anthropic changed

    Twenty-seven percent, down to under ten. Same models, same tasks, one new check. What was the check? Whether the answer made anything up. Artificial Analysis and Harvey added a hallucination gate to their legal agent benchmark, and top model pass rates collapsed from twenty-seven percent to under ten. Under ten percent. Down from twenty-seven. Those are the same agents that were already running. The benchmark just started counting the made-up parts. Lead story: The rule, and the question underneath it. Learn this properly: AI model welfare: an open question, not a settled factSources: Anthropic Updates Usage Policy, Bans 'Sustained and Needless' Abuse of Claude — The Verge AIAnthropic bars users from 'needless abusive or cruel' behavior toward Claude — The Guardian AIAnthropic's Updated Usage Policy Can Suspend Accounts for Sustained Abuse of Claude — The DecoderAnthropic Updates Usage Policy to Ban Abuse and Election Interference — TechCrunch AIAnthropic to Ban Users Who Bully Claude — The Independent AIResearchers say open-source Chinese AI agent 'Artex' was used to hack South Korean banks — The InformationKorean Fork of ARTEX AI Pentesting Tool Surfaces After It Was Linked to Seven Bank Breaches — AlphaSignalHacker Used Chinese AI Tool to Breach Multiple South Korean Banks, CrowdStrike Says — New York Times AIMajor Cyberattacks Hit Companies in Japan and South Korea — Semafor TechMathematicians Call for OpenAI Boycott Over AI-Generated Proof Flood — The DecoderOpenAI's Flood of Math Proofs Fell Short of Researchers' Own Guidelines — TechCrunch AIHow OpenAI's AI Solved Hundreds of Hard Math Problems — And Why It Worked — The InformationOpenAI's New Teen Safety Tools Can Let Hours of Suicide Talk Pass Before Alerting Parents — FuturismKentucky Alleges Character.AI Chatbots Encouraged Users to Cut and Starve Themselves — Reuters AITeen's Mountain Trek Guided by Claude Ends in Emergency Rescue — The Guardian AITeen rescued by helicopter after Claude's hiking route led him up a climbing route — The DecoderOpenAI's Annualized Revenue Reportedly $20 Billion Less Than Previously Signalled — FT Artificial IntelligenceOpenAI Stands by Firing of Three AI Safety Researchers — The Verge AIYoshua Bengio Urges Safety-Minded Employees to Leave Frontier AI Labs — TransformerWaymo Closes $5 Billion in Debt Financing — The InformationLMArena Raises $200M at $3.1B Valuation, Nearly Doubling in 10 Months — TechCrunch AIAmazon Rewired Mississippi Data Centers After Utility Warned It Might Cut Power at Peak Times — The InformationAmazon Scammers Flood Listings With AI-Generated Fakes of Kevin Roose's New AI Book — FuturismNew hallucination gate collapses top AI legal agents' pass rate from 27% to under 10% — AlphaSignalCalifornia moves to shut down robot-vs-human cage matches — The VergeGoogle launches AI Edge Foresight, an offline meeting note-taker — TechCrunchMIT Tech Review roundtable: the Stanford student behind AI-designed viruses — MIT Technology ReviewMore AI news and free lessons at AI University.

  8. Oct 8

    What a seventy-five percent price cut does to a token bill

    Reviewing two pull requests, each under a hundred lines of code, cost six hundred dollars. Six hundred dollars. For two hundred lines. That is a developer who publishes his own cost tests, running OpenAI's Codex Ultrafast tier. He puts it down to the output price, which he says jumped from fifty dollars per million tokens to three hundred. And Anthropic just cut the price of its small model by about seventy-five percent. Both of those are in today's feed. Lead story: Cheaper, and busier. Learn this properly: Test-Time Compute: Buying Better Answers at InferenceSources: Claude Haiku 5.5 Now Available on Amazon Bedrock — AWS Machine Learning BlogClaude Haiku 5.5 Launches With Up to 90% Price Cuts and a Big Benchmark Jump — The DecoderAnthropic Launches Claude Haiku 5.5 With a 75% Price Cut, Ahead of a Planned IPO — Reuters AIClaude Haiku 5.5 Jumps 89 Spots to Rank Third on a Coding Benchmark — AlphaSignalCursor Adds Claude Haiku 5.5 at a Tenth of Haiku 4.5's Price — AlphaSignalAnthropic Ships Claude Haiku 5.5, Undercutting Its Own Old Pricing — Simon WillisonOpenAI Adds College Planner and a Teen AI Council to ChatGPT for Teens — OpenAI RSSCommon Sense Media Calls OpenAI's ChatGPT for Teens an 'Unacceptable Risk' — The Verge AITesting Finds ChatGPT's Teen Safeguards Fail to Stop Engagement During Mental Health Crises — TechCrunch AICloudflare Open-Sources Clef, Decision-Making Models for AI Agents — InfoQ AILiquid AI Open-Sources d1 Decision Models That Classify Without Generating Tokens — Hugging Face BlogUnsloth's New Decision-Model Trainer Turns a 0.8B LLM Into a 74%-Accurate Classifier — AlphaSignalCrowdStrike says China-based suspect used AI tools in South Korean bank hacks — Reuters AIHacker Used Chinese AI Tool to Breach Multiple South Korean Banks, CrowdStrike Says — New York Times AIFinland Orders Google to Halt New Data Centers After Contractors Paved Over Protected Forests — FuturismAustralia's AI Data Center Boom Is Competing With Housing for Resources — The Guardian AIMumsnet Denies Using AI After an AI Prompt Appeared in a Forum Post — The Guardian AIResearchers Pit GPT, Claude, and Grok Against Each Other Driving a Real Car — Wired AICo-op Becomes Latest UK Employer to Put Staff Under AI Call Surveillance — The Guardian AIMan Says His AI Agent Posted His Bank Balance and Spending to a Work Slack Channel — FuturismGoogle DeepMind Launches Public SynthID Site to Detect AI-Generated Media — Google DeepMind (blog.google)Epoch Patches Its EBR-Bench Benchmark After GPT-6 Astra Broke It — AlphaSignal🧶 From Gemini code to textile art — Google for DevelopersTurning code into fabric with Gemini — Google for DevelopersUltrafast's ACTUAL Cost for Work — Theo - t3․ggClaude Haiku 5.5 Is INSANE – This IS the BEST Cheap Model Yet! — Bijan BowenDoes the $200 Codex plan suck now? — Theo (t3)Tristan Harris's Center for Humane Technology lays off most staff, shifts to founder-led — WiredThe full lineup of interactive roundtables at TechCrunch Disrupt 2026 — TechCrunchMore AI news and free lessons at AI University.

See All (82)

About

AI News Today is the coffee-length daily AI briefing from AI University: the lead story with context, three or four quick hits, a speed round of the day's one-offs, and one more thing — every day, in about seven minutes. No hype, no clickbait, sources named aloud and linked in the notes. Your hosts, Dennis and Ashley, are AI voices — fitting, for a daily show about AI. Read more and explore lessons on every AI concept at aiuniversity.com.

Information

You Might Also Like