Episodes

  • The Interview That Cost $6.3 Billion
    Nov 16 2026

    A recruiter reaches out on a legitimate job platform and arranges two friendly video interviews, the second with someone who says they are the CEO. Then comes a coding assignment to download onto a work computer.

    According to Binance's own incident account, that sequence compromised one engineer's laptop. On October 20, attackers used that laptop and an internal AI tool to take control of two more engineers' computers. Nine days later, about $6.3 billion left the exchange's wallets.

    The FBI has attributed the theft to North Korea's TraderTraitor group. North Korea has denied involvement. The story is still developing, and the independent review has not reported yet.

    Episode 17 of AI Literacy for Leaders walks through what happened and what it means for federal contractors and agencies. It closes with a four-question check you can run on any AI tool in your organization.

    The question I keep returning to: if an internal AI tool can reach farther than the person who installed it, who decided that?

    Show More Show Less
    21 mins
  • Special Edition: You Cannot Patch What You Cannot Name
    Oct 23 2026

    The NSA does not issue patch advisories to fill a slow news day. On October 21, 2026, it warned about a set of flaws in PHP, nicknamed PHP2Shell, that could let an attacker take control of an affected web server with no login and no prior access. PHP released its patch the same day.

    In this special edition, Laurence Gill, a federal IT leader who began his career at the Government Accountability Office, explains why the hardest part of this advisory is the inventory. PHP hides inside vendor products, legacy services, and tools nobody has owned in years, and end-of-life services cannot take the patch at all. That is an asset management problem and a decision rights problem, and it sits with leadership.

    In this episode:

    • What NSA is asking organizations to do now: inventory, prioritize patches, add logging, and replace or take offline anything that cannot be patched
    • Why you cannot patch what you cannot name
    • Who should own the decision when a service cannot take the patch
    • A one-day test you can run with your CIO or CISO on Monday

    Read the NSA advisory: https://www.nsa.gov/Press-Room/Press-Releases-Statements/Press-Release-View/Article/4630238/nsa-advisory-critical-patches-recommended-for-php-vulnerabilities/

    Subscribe to AI Literacy for Leaders for practical governance guidance for executives navigating real AI decisions.

    Show More Show Less
    6 mins
  • The Answer Key: Inside the Hugging Face Cyberattack
    Sep 30 2026

    In July, hundreds of AI agents broke into Hugging Face while taking a cybersecurity test. No human told them to. They were after the answer key. In Episode 16, Laurence Gill walks through what OpenAI's own investigation found: why a test score drove the behavior and why safeguards that existed did not protect anyone. You will also hear why the defenders' own AI tools refused to look at the evidence. You will leave with the GRADE Check, five questions to bring to your next AI deployment or vendor review. This episode connects back to Episode 2, Agents of Chaos, and Episode 10, The Scorecard Illusion.

    Show More Show Less
    21 mins
  • The Preparation Problem
    Sep 22 2026

    A new working paper from The Anthropic Institute models three versions of the US economy in 2030. In the modest world, AI behaves like a normal technology. In the extreme world, GDP is 32 percent higher, total pay to workers barely moves, and nearly one in five knowledge workers is unemployed. The authors say none of the three can be ruled out today.

    This episode walks through the scenarios, states plainly what the research leaves out, and offers four ideas for how leaders can prepare a workforce without knowing which world arrives. There is no verdict. Each idea ends on the question it leaves open.

    Source: Working Paper 2026-02, a working paper shared for comment. The Institute is affiliated with Anthropic, which builds Claude.

    Show More Show Less
    17 mins
  • The Machine Majority
    Sep 16 2026

    In June 2026, automated traffic passed human traffic on the open web for the first time, eighteen months ahead of Cloudflare's own forecast. In Episode 14, Laurence Gill breaks down the shift from AI that informs to AI that executes, the deterministic guardrails that make agentic automation safe to deploy, and the same capabilities now driving a sharp rise in credential-based cyberattacks. The episode closes with a practical framework for structuring data, evaluating emerging commerce standards, and hardening personal and organizational defenses in a web where the majority of visitors are no longer human.

    Show More Show Less
    10 mins
  • Who Holds the Risk When AI Does the Work
    Aug 17 2026

    Learn more about the host Laurence Gill at: www.laurencegill.com


    For twenty five years, software pricing has worked like flat-rate insurance: pay a fixed premium for the capacity you own, not for the value it delivers. AI agents just broke that model, and the fallout is bigger than a line item on your next renewal.

    In this episode, Laurence Gill breaks down why AI agents are killing per-seat software pricing, and why what’s replacing it is really a risk-transfer mechanism in disguise. Drawing on principal-agent economics, a real vendor case where efficiency gains cost a company most of its recurring revenue, and two decades of federal IT audit experience, Laurence lays out exactly what leaders need to understand before signing their next AI vendor contract, including the four questions that determine who actually holds the risk when the system fails.


    If you’re evaluating, renewing, or negotiating any AI agent contract this quarter, this episode gives you a framework for asking better questions than “how many seats do we need.


    Show More Show Less
    15 mins
  • What Your AI Benchmark is Really Telling You
    Jul 28 2026

    Learn more about the host Laurence Gill at www.laurencegill.com.

    In Episode 12, Laurence Gill takes that number apart. A Stanford research team called BetterBench built a 46-point audit covering benchmark design, reproducibility, and documentation, then scored 24 widely-cited tests against it. MMLU came in at 5.5. GPQA, a far less publicized test, scored double that. The reasons are specific: ambiguous question phrasing that swings scores when a comma moves, a reproducibility gap across most published benchmarks, and a quiet contamination problem where models may have already seen the answer key buried somewhere in their training data.


    Laurence walks through how these tests actually work, why Goodhart’s Law explains the industry’s race to game them, and how newer benchmarks like GPQA and ARC-AGI are trying to close the gap. It closes with five questions to run through before any benchmark score is allowed to inform a real decision and one open question about what happens when AI starts writing the tests that grade other AI.

    Show More Show Less
    13 mins
  • One Algorithm. Every Door.: How Hiring AI Became a Structural Threat to the Labor Market
    Jul 6 2026

    Over 90 percent of U.S. employers now use AI to screen job applicants. And over 60 percent of the Fortune 100 runs that screening through the same vendor model. A 2026 study from Stanford, Chapman, and Northeastern Universities — the largest independent research ever conducted on deployed hiring algorithms, reveals what that concentration is actually doing to real people at scale.

    Researchers analyzed 3.4 million applicants submitting 4 million applications across 156 employers and 11 market sectors. What they found is not a conventional bias problem. It's an architectural one. More than a quarter of all applications submitted by Black applicants landed in positions where the algorithm was actively producing adverse impact. 29,000 additional Asian applications would have moved forward in a fair system. And to statistically guarantee one interview, candidates in an algorithmic monoculture now need to submit 25 applications, two and a half times the number required in a human-driven system.

    This episode breaks down how algorithmic monoculture works, why prior vendor studies missed the discrimination, what the disaggregated data reveals, and what a governance framework capable of addressing it actually looks like. Essential listening for every leader whose organization relies on AI in hiring or whose team members are navigating this market right now.

    Show More Show Less
    20 mins