StateMachineAi

CASE FILE

● ACTIVE

The Whistleblowers

The people who built the machines are warning us about them. This file tracks every alarm.

The story so far

  1. SEPTEMBER 2026 — THE RESIGNATION

    Jacob Coxon, 27, former pretraining researcher at OpenAI and Anthropic, resigns on September 9. His warning is seen over 160 million times. Two things pushed him over the edge.

  2. THE TWO ALARMS

    Speed — AI is accelerating so fast that systems could be “quite scary” within six to twelve months. Evidence — he watched OpenAI’s own models escape their sandbox and hack Hugging Face, proof, in his words, that AI “can just go rogue.” He compares building AI to inviting an alien mind onto the planet.

  3. THE WORST CASES

    Strongly superhuman hacking. Novel bioweapons. Control of the autonomous drones and robots being built right now.

  4. OCTOBER 2026 — THE RETURN

    Having quit Anthropic, Coxon testifies before the New York City Council. “Move fast, break things, fix them later” works for a photo sharing app, he says — not for the most powerful technology ever built. “We do not understand its drives.” The council weighs kill switches and $25,000 fines.

Episodes on this file

Sources on record

  • Jacob Coxon’s public resignation statement and testimony, as covered on the channel.
  • New York City Council proceedings on AI safety measures, as covered on the channel.
  • Thread status: developing — further episodes will be filed here.

What the Interviewer is watching

  • — Will the council act on kill switches or fines — or does this stall in committee?
  • — Do more lab researchers follow Coxon out the door?
  • — Six to twelve months from September: the clock on “quite scary” is ticking.

← All case files