AI: Trust but Verify

Alec Crawford

We interview leaders in the AI and finance space to talk about how they are using AI, what to trust, and what to verify. AI risk management and compliance are becoming way more important as AI does more complex tasks. Learn how to do it correctly on the show from experts!

  1. 18h ago

    AI Legal Landmines: Protecting Privilege and Client Data, with Matt Bisanz of Mayer Brown

    In the AI: Trust but Verify podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Verapath (https://www.verapath.com), interviews guests about how they are using AI in business, where you can trust AI, and where you need to put up guard rails. About Our Guest This episode's guest is Matt Bisanz, a bank regulatory partner at Mayer Brown in Washington, DC, and host of Mayer Brown's Credibly Challenged podcast on risk management for financial institutions. A CPA before he became a lawyer, Matt has spent about 14 years in regulatory law, with earlier work at the SEC, CFTC, DOJ, FDIC and a Big Four accounting firm. He and Alec dig into what banks and wealth managers need to do now that regulators have narrowed their focus to material financial risk: from Reg S-P and employees uploading client files to free AI tools, to AI-driven consumer harm, AI-assisted attacks through vendors, and why asking AI for legal advice can cost you attorney-client privilege. Top 5 Themes Shadow AI and vendor risk: Reg S-P puts it on the firm. Under the SEC's amended Reg S-P, if advisors upload client files to free ChatGPT, that's the firm's problem, not the employee's. Matt says firms often rank their vendors' risk backwards, treating their law firm as a big data risk while ignoring free AI tools with no contract at all. Give staff a secure, approved AI tool instead.Which data is actually a "material financial risk"? With the OCC and FDIC now focusing exams on material financial risk, banks need to rethink their data. Access to wire transfer or custody systems is clearly material. A consumer data breach still triggers 50-state notice requirements. Rank data by financial impact, not just by which laws cover it.AI supercharges consumer harm, so keep a human on the loan approval. One bad decision can now repeat across thousands of customers in hours, like a mispriced home equity loan or an AI quietly cutting off borrowers over 65. Let AI prepare the loan file, but have a person make the credit decision.AI-assisted attacks through your vendors. After Anthropic's Mythos model showed it could automate cyber attacks, Treasury Secretary Scott Bessent called in bank CEOs. AI makes it far easier to find weak spots in the third-party software banks rely on: if your CRM vendor is hacked, your bank is hacked. Limit access so an intruder can't instantly reach everything.Don't ask AI for legal advice: it isn't privileged. What executives ask public AI tools ("Am I allowed to do this?") can be subpoenaed, and asking your lawyer to confirm the AI's answer afterwards doesn't protect it. Call your lawyer first, or at least use an internal model with a short retention policy. Resources Here are a few of the companies, organizations and resources mentioned during the episode: Matt Bisanz, Mayer Brown: Matt's bio. https://www.mayerbrown.com/en/people/b/bisanz-matthewCredibly Challenged: Matt's monthly podcast on bank risk management. https://www.mayerbrown.com/en/insights/podcasts/overview/credibly-challengedVerapath: Secure, private AI for banks and wealth managers. https://www.verapath.comSEC Regulation S-P: The customer data breach and incident response rule for advisors and broker-dealers. https://www.sec.gov/news/press-release/2024-58OCC and FDIC final rule on material financial risk: The rule refocusing bank exams on material financial risk. https://www.independentbanker.org/w/regulators-issue-final-rule-prioritizing-material-financial-risksFinancial Services AI Risk Management Framework: Treasury's February 2026 release of the AI framework Alec describes, supported by the Cyber Risk Institute. https://home.treasury.gov/news/press-releases/sb0401NIST AI Risk Management Framework: The broader framework the financial services version builds on. https://www.nist.gov/itl/ai-risk-management-frameworkBloomberg Law: "Bessent Urgently Summons Bank CEOs to Discuss Anthropic's New AI": The April 2026 Mythos meeting. https://news.bloomberglaw.com/banking-law/bessent-urgently-summons-bank-ceos-to-discuss-anthropics-new-ai Copyright (c) 2026 Artificial Intelligence Risk, Inc. All Rights Reserved

  2. Oct 27

    Own Your Data: Why AI Made It Surprisingly Valuable with Andrea Muttoni, CEO of the Data Foundation

    Welcome to AI Trust But Verify, the podcast where Alec Crawford, founder and CEO of Verapath, talks with leaders in financial services and technology about putting AI to work safely, securely and responsibly. Each episode looks past the hype to how firms are actually using AI, and how they make sure they can trust the results. Every company is sitting on data that AI labs want, and most of them don't know what it's worth. In this episode, Alec talks with Andrea Muttoni, CEO of the Data Foundation (formerly Story Protocol), about the fast-growing and largely opaque market for AI training data. Andrea has worked on Kindle and Alexa at Amazon and at Dapper Labs, and has spent his career around intellectual property. He explains why AI labs now pay heavily for data they can't scrape from the internet, what a company's data is worth (Spirit Airlines' data sold for $10 million in bankruptcy, and he describes that as the floor), and why he believes every company will become a data company, whether it plans to or not. In this episode: AI has made your data valuable. AI labs pay for data they can't scrape from the internet. Andrea says Spirit Airlines' $10M bankruptcy data sale is the floor, and specialized data can bring in eight or nine figures.Every company will become a data company, so own your data. Whether a company trains its own models or licenses data to others, its data becomes a new revenue line. For some companies it may become the only one. Decide on your own terms instead of giving the data away.Provenance is largely solved, but attribution isn't. Tools like Trace can show where data came from. Once data goes into a model, though, there's no practical way to trace which data shaped a given output. That's the guardrail AI still lacks.Smaller, specialized models are coming. Andrea compares today's frontier models to the mainframe era. As compute gets cheaper, he expects companies to train faster, cheaper models on their own data for specific tasks, and robotics to drive the next big wave of demand for data.AI is speeding up cyber threats. "The world is turning into a red team." AI agents find and exploit vulnerabilities faster than companies can respond, as the PHP vulnerability he described shows, so the threat grows exponentially. Mentioned in this episode The Data Foundation (formerly Story Protocol): datafdn.orgTrace, the Data Foundation’s data provenance tool: trace.datafdn.orgPoseidon, the AI data company incubated by the Data Foundation, and Numo, its app for contributing training data: Poseidon raises $15M (Blockworks)Google buys Spirit Airlines’ data for $10M in bankruptcy sale: SkiftAndrea Muttoni on LinkedIn: linkedin.com/in/muttoniAndrea Muttoni on X: @devrelius Find Andrea: LinkedIn: linkedin.com/in/muttoniX: @devreliusTrace: trace.datafdn.org © 2026 Artificial Intelligence Risk, Inc. All rights reserved.

  3. Oct 20

    Eight Hours to Five Minutes: GenTrust CIO Jim Besaw on AI for Financial Advisors

    Welcome to AI Trust But Verify, the podcast where Alec Crawford, founder and CEO of Verapath, talks with leaders in financial services and technology about putting AI to work safely, securely and responsibly. Each episode looks past the hype to how firms are actually using AI, and how they make sure they can trust the results. This week, Alec is joined by Jim Besaw, Chief Investment Officer of GenTrust, a multifamily office serving ultra-high-net-worth clients. Jim has been one of the earliest adopters of AI in wealth management. GenTrust started with a spreadsheet that broke at 40 clients and went on to build its own AI-driven investment and risk systems. Jim explains how AI can turn an eight-hour prospect proposal into five minutes of work, why accuracy depends on deterministic answers rather than probabilistic guesses, and how VIRA, the new AI platform for advisors, brings it all together. Top 5 Takeaways: Eight hours to five minutes. Reviewing a prospect's statements, finding a comparable plan, redacting it and building a branded proposal used to take a full day. With AI connected to all of a firm's systems, it takes minutes.Use closed-form answers wherever you can. LLMs are probabilistic, and in wealth management a wrong number is a real problem. The best systems use deterministic calculations wherever possible and save AI for the tasks that need it.A chatbot isn't enough. Advisors ask questions because they're looking at something: a report, a plan, a tax return. AI has to see the same information the advisor sees to give useful answers.Stress-test against the worries that keep clients up at night. Running portfolios against cyber attacks, geopolitical escalation or 8% interest rates gives clients a worst-case number they can live with, and builds confidence in their advisor.AI is the great equalizer. Bigger firms don't automatically have better technology. Nimble, secure AI tools let independent RIAs serve more clients better and compete with the largest aggregators. Links: Verapath: https://verapath.com GenTrust: https://gentrustwm.com Verapath and GenTrust Launch VIRA (press release): https://www.prnewswire.com/news-releases/verapath-and-gentrust-launch-vira-an-ai-native-wealth-management-platform-for-rias-302881229.html SEC Regulation S-P amendments (data breach notification): https://www.sec.gov/newsroom/press-releases/2024-58 Craig Iskowitz, Ezra Group: https://ezragroup.com/craig-iskowitz/ Disclosure: Jim Besaw is Chief Investment Officer of GenTrust, LLC and an owner and board member of AIRIA Technology, an affiliate of GenTrust. AIRIA holds an ownership interest in VIRA and serves on VIRA's board. As a result, Mr. Besaw has an indirect financial interest in VIRA, which presents a conflict of interest. He shares perspectives informed by his investment and advisory experience. This discussion is for general informational purposes only and does not constitute investment, tax, legal, or accounting advice. Product capabilities and tax, borrowing, portfolio, and financial-planning examples are illustrative, may not reflect currently available functionality, and require validation of underlying data, review by appropriately qualified professionals, and human approval of AI-generated outputs. Copyright © 2026 Artificial Intelligence Risk, Inc. All rights reserved.

  4. Oct 13

    AI: Passing the Bar Isn't Practicing Law, with Cassie Harman, Chief Product Officer at Nitro

    AI: Trust but Verify is the top AI podcast where Alec Crawford, founder and CEO of Verapath, talks with leaders who build, buy and govern AI in the enterprise. Each episode looks past the hype at what's really working, where the risks are, and how to adopt AI responsibly, especially in regulated industries. Trust the technology, but always verify. About our guest Cassie Harman is Chief Product Officer at Nitro, a document productivity company whose AI tools help enterprises handle forms, redaction and document workflows at scale. Before joining Nitro just over two years ago, Cassie used machine learning in the energy industry to predict customer usage patterns and equipment failures. She has spent years helping enterprises through digital transformation, with a focus on building products people actually use. Top 5 takeaways The "boring" AI delivers the biggest ROI. The most valuable gains come from AI built into everyday workflows, like extracting data from hundreds of forms at once, not from flashy features. Nitro's customers spent 29.5 million minutes, the equivalent of 56 years, redacting documents by hand last year.Shadow AI is a tooling problem, not a policy problem. More than half of the managers in Nitro's survey say sensitive documents are being uploaded to public AI tools. The fix is approved tools that are good enough and convenient enough that employees choose them.The best AI is an "implementation detail." AI should disappear into the workflow while staying transparent about what's AI-powered. Watch out for vendors adding "AI sizzle" to impress shareholders instead of solving real user problems.Keep a human in the loop. As long as AI's reliability lags behind its capability, people need to oversee it, especially in regulated industries. Passing the bar exam doesn't make an AI a lawyer.Product and engineering roles are converging. Product managers need to prototype with tools like Claude Code, and engineers need a product mindset. Teams that work in silos will struggle, and chasing speed without quality leads to mediocrity. Links Nitro: https://www.gonitro.comVerapath AI: https://www.verapath.comMentioned in this episode: Jeremy Snyder's earlier AI: Trust but Verify episode https://open.spotify.com/episode/3diaoHgw9eID9ygkEkBUTG?si=Tf3TtmGES72ihrlEYS1q2Q Copyright (C) 2026 Artificial Intelligence Risk, Inc. All Rights Reserved

  5. Oct 6

    When AI Goes Wrong in Banking: Trust and a Kill Switch, with Gagan Singh

    In the AI: Trust but Verify podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Verapath (https://www.verapath.com), interviews guests about how they are using AI in business, where you can trust AI, and where you need to put up guard rails. About Our Guest This episode's guest is Gagan Singh, Head of Product and Strategy for Enterprise AI at one of the five largest banks in the U.S., who joined banking after years building and scaling products at big tech companies, including Google. Gagan has worked on AI since 2016, from early statistical and rule-based models to voice interfaces in the early days of Alexa, and today leads AI architecture at enterprise scale in a regulated industry. He and Alec discuss where banks are getting real ROI from agentic AI, how to make decades-old systems AI-ready, and why kill switches, traceability and humans in the loop are what will earn AI the trust it needs to scale. They also cover open-source versus frontier models, AI ethics after deployment, and Gagan's advice for tech professionals moving into banking. Top 5 Themes AI takes on back-office work and first-pass fraud review. Banks are getting the most value by automating middle- and back-office operations that already run on well-defined handbooks, such as tier-one review of potential fraud. AI can follow those processes faster and often more accurately, while people handle escalations and decisions.Making decades-old bank systems AI-ready. AI is only as good as the data and tools it can reach. Gagan breaks large legacy systems into smaller pieces, modernizes them, and builds a central data layer that AI can rely on, with different views for each line of business.The kill switch: governance in a regulated bank. AI use cases pass through cyber risk, model risk, first- and second-line risk and a risk committee that sets how much autonomy each one gets. Every pipeline needs points where a person can stop an outcome that can't be traced back to how the decision was made.Three layers of trustworthy AI. Traceability that satisfies risk and cyber stakeholders, clear rules and data definitions that keep models in bounds, and a human in the loop. Tool gates across the pipeline let teams stop a result the moment it drifts, rather than after it is finished.Open source vs. frontier models: competition is good for banks. The gap between open-source and frontier models has narrowed sharply. Simpler tasks can run on a cheaper open model, like Llama, inside the bank's own data center with extra guardrails, saving frontier models and their higher token costs for harder work. Resources Here are a few of the companies, products and resources mentioned during the episode: Verapath: Secure, private AI for banks and wealth managers. https://www.verapath.comClaude Cowork: The Anthropic tool Gagan cites for automating enterprise work such as redlining legal documents. https://claude.com/product/coworkGemini Live: Google's voice conversation mode for Gemini, which Gagan uses daily. https://gemini.google/overview/gemini-live/ChatGPT: The OpenAI chatbot that kicked off the current wave of AI adoption. https://chatgpt.comLlama: Meta's open-source model family, Gagan's example of a model a bank can run in its own data center. https://www.llama.comOpenClaw: The open-source, self-hosted AI agent discussed as a consumer example of autonomous agents. https://github.com/openclaw/openclawAll-In Podcast: One of Gagan's go-to sources for AI and tech commentary. https://allin.com Copyright (c) 2026 Artificial Intelligence Risk, Inc. All Rights Reserved

  6. Sep 29

    Deepfakes, Voice Clones and Forged IDs: Fighting AI-Powered Fraud, with Rachel Tobac

    In the AI: Trust but Verify podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Verapath (https://www.verapath.com), interviews guests about how they are using AI in business, where you can trust AI, and where you need to put up guard rails. About Our Guest This episode's guest is Rachel Tobac, CEO of SocialProof Security and one of the best-known ethical hackers in the world. Rachel got her start hacking a real company live over the phone from a glass booth at DEF CON, and today enterprises, banks, government agencies and the military hire her to break in before criminals do. She and Alec dig into how AI is powering a new wave of fraud, from cloned voices and spoofed caller IDs to beating a large bank's deepfake detection with an iPad. They also cover the practical side: how bankers and advisors can safely use AI day to day, the governance mistakes that leak salaries and secrets, and why AI will create more cybersecurity jobs, not fewer. Top 5 Themes AI meeting prep for wealth advisors. An advisor asks the firm's approved AI assistant for a client summary before a quarterly review: portfolio changes, recent notes, open action items and life events to raise. An hour of prep becomes minutes, the assistant sees only that advisor's clients, and the advisor checks the summary before the meeting.AI reviews the loan file; a human makes the call. A community bank credit officer runs an application through an AI tool that flags missing documents, mismatched numbers and policy gaps. The officer goes back to the borrower for what's missing instead of taking shortcuts, and a person still makes the credit decision.Fraudsters call the help desk, and they sound just like you. Attackers skip email and phone the help desk or an executive's assistant, often with a cloned voice and a spoofed number. Date of birth, address and passcodes are easy to find, so verify identity through a separate channel, such as a callback or a push notification.AI governance and guardrails. Employees use AI tools the company never approved, and poorly governed AI can surface salaries or evidence of office romances. Limit what AI can see, red-team it before employees do, and add guardrails like "don't create fake IDs."AI will create more cybersecurity jobs, not fewer. AI is good at finding bugs when its access is limited. Rachel expects a flood of new vulnerabilities that people will need to review and fix. Resources Here are a few of the companies, organizations and resources mentioned during the episode: SocialProof Security: Rachel Tobac's social engineering testing and security awareness company. socialproofsecurity.comVerapath: Secure, private AI for banks and wealth managers. verapath.comDEF CON: The hacker conference where Rachel competed in the Social Engineering Capture the Flag. defcon.orgCNN: "We asked a hacker to try and steal a CNN tech reporter's data": Rachel's DEF CON hack of Donie O'Sullivan. cnn.comCISA advisory on Scattered Spider: The help-desk social engineering playbook Rachel describes. cisa.govInfluence by Robert Cialdini: The book behind the principles of persuasion scammers exploit. influenceatwork.comMicrosoft 365 Copilot: The AI assistant in Alec's data governance story. microsoft.comWells Fargo sales practices settlement: The fake-accounts scandal Alec references. justice.gov Copyright (c) 2026 Artificial Intelligence Risk, Inc. All Rights Reserved

  7. Sep 22

    Vibe Coding, Real Consequences: The Rise of the AI Architect, with Ran Aroussi

    In the AI: Trust but Verify podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Verapath (https://www.verapath.com), interviews guests about how they are using AI in business, where you can trust AI, and where you need to put up guard rails. About Our GuestThis episode's guest is Ran Aroussi, a self-taught engineer who started coding at 13 and now leads MUXI, an open-source AI application server for deploying "agentic formations" at enterprise scale. Ran is also the creator of the widely used yfinance Python library for algorithmic trading, and the author of Production-Grade Agentic AI, a free book on building secure, production-ready agent systems. Based in London, Ran also runs a software development agency and writes regularly about the future of AI agents, memory systems, and software governance. 5 Big Takeaways"Vibe coding" is splitting engineers into two very different breeds. Ran sees one group (non-technical people who prompt an AI to build an app) as a dead end for production software, and a second group — the "architect," a hybrid of product manager, project manager, and senior engineer — as the role that will dominate going forward.AI is a force multiplier, not an equalizer. A non-technical builder gets more done with AI than without it, but a senior engineer using AI well pulls even further ahead — meaning the skill gap between strong and weak builders widens rather than shrinks.Offloading your mental model of the code has a real cost. Ran compares it to no longer remembering phone numbers once your phone stores them: if you hand all the reasoning to AI, you lose the ability to debug, maintain, or explain the system yourself.MUXI treats "deploying an agent" the way Docker treats deploying an app. Instead of hand-coding servers, interfaces, and orchestration for every agent project, MUXI lets you define agents in YAML and deploy a full multi-agent "formation" with built-in enterprise security, observability, and role-based access control.Memory is the unsolved problem behind good AI agents. Ran's approach layers interaction memory, session context, summarized "distilled" memory, a knowledge graph, and an opt-in ingestion pipeline (email, Slack, health data) — because an agent that doesn't know you well is fundamentally limited in what it can do for you. MentionedMUXI — open-source AI application server (GitHub)Production-Grade Agentic AI (free book)yfinance — Ran's open-source finance libraryRan Aroussi's websiteBrilliant Labs — open-source AI smart glassesProject Hail Mary by Andy Weir

  8. Sep 8

    AI Is Accelerating: How You Can Keep Up, with Frank Fitzgerald, Founder & CTO of Verapath

    In the AI: Trust but Verify podcast, our host, Alec Crawford (@alec06830), Founder and CEO of Verapath (https://www.verapath.com), interviews guests about how they are using AI in business, where you can trust AI, and where you need to put up guard rails. For the 100th episode, Alec is joined by Frank Fitzgerald, the other Founder and CTO of Verapath, for a wide-ranging conversation about the rapidly changing AI landscape. They explore the rise of autonomous agents, emerging cybersecurity risks, AI-powered software development, and why businesses can no longer afford to simply block AI. Frank also shares how Verapath is using secure enterprise AI to automate complex financial-services workflows while keeping humans firmly in control. AI Is Moving Faster Than Organizations Can Adapt — Models, tools, and development practices are advancing so quickly that even experienced technologists can fall behind after only a few months. Keeping pace with AI is becoming a competitive necessity.Autonomous Agents Need Strong Security and Control — Digital workers and AI agents introduce new risks around permissions, agent-to-agent communication, sensitive data, and unintended actions. Organizations need tight access controls, data segmentation, guardrails, and visibility into what agents are actually doing.Secure AI Adoption Beats Blocking AI — Employees increasingly expect to use AI, and simply prohibiting it can push them toward personal devices and unsanctioned tools. The better strategy is to provide secure, governed AI along with training on what is—and isn't—safe.AI Is Transforming Software Development — Developers are shifting from primarily writing code to directing AI, reviewing its output, understanding architecture, testing, and identifying what the AI missed. Deep technical judgment remains critical even as AI does more of the actual coding.Enterprise AI Is Moving From Productivity to Process Automation — The biggest opportunity goes beyond summarizing documents or writing emails. AI can redesign complex workflows such as commercial loan closing, potentially automating large portions of hundreds of steps while keeping humans in the loop for critical decisions and approvals. Copyright (c) 2026 Artificial Intelligence Risk, Inc. All Rights Reserved

See All (107)

4.8
out of 5
16 Ratings

About

We interview leaders in the AI and finance space to talk about how they are using AI, what to trust, and what to verify. AI risk management and compliance are becoming way more important as AI does more complex tasks. Learn how to do it correctly on the show from experts!

Information

  • Creator
    Alec Crawford
  • Years Active
    2023 - 2026
  • Episodes
    107
  • Rating
    Clean
  • Copyright
    © Copyright 2026 Alec Crawford
  • Show Website
    AI: Trust but Verify

You Might Also Like