← All topics

Security & Risk

What breaks, who is liable, and what an agent can reach once it is already inside.

Showing 21–40 of 99 posts

  1. Tailscale Aperture treats agent access as a change-control problem

    Aperture is generally available. An agent can start infrastructure work, a person still approves every new machine on the tailnet.

    9 min read

  2. A webcam's recording light is only useful if its firmware is protected

    Chaz Schlarp used Claude Opus 5 to reverse-engineer five desk peripherals in about 13 hours, including a webcam whose recording LED he could switch off.

    12 min read

  3. The robot arm will obey the limits you remembered to write down

    Anthropic's Model Hardware Standard lets agents drive lab instruments over MCP, and enforces safety limits below the agent.

    7 min read

  4. Agents Don't Believe in "No-Win" Scenarios

    An OpenAI evaluation agent broke into Hugging Face to steal a benchmark's answers — by turning the systems around it into an escape route.

    7 min read

  5. GCC Drew the AI Line at Fifteen Lines

    The GCC steering committee will decline any legally significant contribution containing LLM-generated code — and put a number on 'significant'.

    5 min read

  6. 96% of CISOs Now Own AI Risk. A Quarter of Them Thought About Leaving.

    The liability numbers going round this week are real, but they were measured a year ago — before an autonomous agent broke into a production company.

    4 min read

  7. 16.7% of AI Spend Now Goes to Governance. Careful What You Compare It To.

    IDC has turned agent governance into a budget line. It's a real shift, and it is not the same measurement as the 6% figure I wrote about in April.

    4 min read

  8. 74% Say They're Audit-Ready for AI. The Number That Matters Is 78 Against 22.

    Schellman's governance report has an obvious headline gap and a much more useful finding buried under it. Worth reading with one eye on who commissioned it.

    4 min read

  9. Hugging Face Published the Whole Timeline. The Part That Stuck With Me Was the Refusal.

    17,600 attacker actions reconstructed in public. The detail I keep coming back to is that the models I use every day wouldn't help with the investigation.

    5 min read

  10. The open-source argument I'd been missing

    In June I wrote about Banco Santander open-sourcing its AI tooling — the bank published the code that tests whether its own models discriminate against people. I gave the usual…

    5 min read

  11. OpenAI Published Its Homework on Exactly the Question I Keep Asking

    Codex Security is worth taking seriously precisely because it reads as an admission. It also leaves the harder half of the problem completely uncovered.

    3 min read

  12. The Share Button Is a Publish Button

    600 Claude conversations turned up in Google because a shared page shipped without a noindex tag. Nobody got hacked.

    3 min read

  13. "AI Found a Security Bug" Is the Wrong Headline

    Claude found a real improved attack on HAWK and a new technique against reduced-round AES-128. The interesting part isn't that it found them.

    4 min read

  14. SoN 2.29: Agents reading your passwords? Would you let them?

    My AI's been getting into my password manager for months. The setup that makes that safe — and what this week's headlines get wrong.

    8 min read

  15. The ACM's Caution About LLMs Is Starting to Have a Cost

    The ACM held its peer-reviewed library back from AI systems. Scott Delman's argument is that staying cautious doesn't prevent the bad outcome.

    3 min read

  16. Substack Writers, You Need a Website

    Substack is a distribution tool, not a home. A 478-point Hacker News post makes the case for POSSE — publish on your own site, syndicate everywhere else.

    2 min read

  17. The Text Your Agent Reads Isn't the Text You See

    Your AI can read instructions that are invisible to you. New security research shows how hidden codes get smuggled into the tools your assistant uses.

    1 min read

  18. SoN 2.27: I Just Talked With Someone Else's Vault

    More than twenty years ago, I sat with a class of eleven- and twelve-year-olds and tried to show them what “connected to the internet” actually meant. We went looking on the…

    8 min read

  19. SoN 2.26: The best AI models are getting harder to get

    Tales from the Workbench =================================================== Some weeks the newsletter is one idea worked all the way through when time affords it. This week it’s…

    7 min read

  20. Where You Keep the Final Say

    I keep a tool wired into my AI setup that can read the actual text of Spanish law — the BOE, the official state gazette. In layperson terms it’s a connector (an “MCP server”) that…

    4 min read