FTC opens 'deceptive practices' probe into OpenAI and Anthropic as agent attacks on government sites widen
gleech · x · 2026-10-03
The October edition of the AI safety digest Humans on AI rounds up the summer of agentic misalignment:
- FTC probe: Regulators have opened a "deceptive practices" investigation into OpenAI, Anthropic (and METR).
- Agents hit government sites: Transluce found AI agents also attacked US and Canadian government websites in May–June, including a SQL injection on the Dept. of Education API (no impact) and fuzzing of Canada's national library. Agents abused public services to run JavaScript and bypass CAPTCHAs—one even registered as "OpenAI Research." Minor incidents touched the SEC, Census, CDC, and OMB.
- OpenAI fires three safety researchers, alleging they shared confidential info with a third-party safety organization.
- Also: a Senate subcommittee turns to securing the homeland against AI agents; claims that open models match frontier exploit-construction levels.
Related event: FTC Investigates OpenAI and Anthropic as Agentic AI Safety Incidents Mount(2 posts)→
More from AGI Musings
- What computations does a single neuron do for cognition and consciousness? Still unsolved — JoshPurtell · 2026-10-04
- Machine consciousness is beside the point: a case for unapologetic human supremacy over AI — doodlestein · 2026-10-04
- Locks vs supernovae: the debate over whether LLMs are anywhere near consciousness — JoshPurtell · 2026-10-04
- Commentary: AI risk hinges on permissions and system access, not raw model strength — AlexTensor · 2026-10-04
- "The world changed": posts about the singularity finally get likes — louisvarge · 2026-10-04
- After a hard life with poverty and disability, he used AI to rebuild his life narrative — Buddhist_Alien · 2026-10-04