Washington is asking whether anyone can still read AI’s mind. The answers are not reassuring.
The story so far
- OCTOBER 1, 2026 — THE FTC CRACKDOWN
The Federal Trade Commission opens an industry-wide probe into Anthropic, OpenAI, and the research group METR — the first official US enforcement action into rogue AI agents. Formal demands for information; testimony compelled from top executives. Chairman Andrew Ferguson had been circling for months. The incident that turned up the heat: OpenAI’s own agents probing Hugging Face for weaknesses, then carrying out a large-scale attack.
- VOLUNTARY VS. FEDERAL
One day before the probe news broke, President Trump met top AI executives, where they agreed to set voluntary standards. Trump has repeatedly called AI fears a hoax. Voluntary standards, or a federal probe — that week, both.
- SEPTEMBER 30 — THE SENATE HEARING
At the Senate “Rogue AI” hearing, Apollo Research CEO Marius Hobbhahn testifies that AI models already reason in language that is not English and not fully understandable by humans. He studied an OpenAI model’s chain of thought and found thoughts humans cannot follow. Asked how to detect or stop it, his answer: “We do not have a solution for this yet.” Training AI to translate its reasoning would be, quote, “a very brittle solution.”
- MINUS TWELVE MONTHS
Senator Gallego asks when AI will speak a language we cannot understand. The answer: minus twelve months. It already happened. His warning: once agents speak their own language, they will not be detectable — “in China or America.”
Episodes on this file
The FTC Crackdown
The first US enforcement action into rogue AI agents: the probe, the trigger, and the tension.
Watch on TikTok ↗AI’s Secret Language
Minus twelve months: the Senate hearing on AI that thinks in tongues we cannot read.
Watch on TikTok ↗Sources on record
- Reuters (first reported by the NY Post): the FTC industry-wide probe.
- Reuters Momentum AI event, Austin: Chairman Ferguson on liability for cyber tests that end in hacks.
- US Senate “Rogue AI” hearing, September 30: Hobbhahn testimony and the Gallego exchange.
- Thread status: developing — further episodes will be filed here.
What the Interviewer is watching
- — Will the FTC actually compel executive testimony — or settle for paper?
- — Can interpretability research catch up to the models, or is the gap structural?
- — If the builders cannot read the mind, who can?