Full Breakdown
AI Researchers Push for Industry-wide Slowdown After High-Profile Resignation
By Drooid · · How we work
The Catalyst: Jacob Coxon’s Exit
Coxon’s resignation post on X warned that the race to “self-improving superintelligence” is a gamble with lives. The outcry sparked private dinners, encrypted chats and a nascent “Coalition of Concerned AI Staff” focused on safety advocacy.
Timeline of Key Events
| Date | Event |
|---|---|
| Sept 8 | Coxon posts resignation and raises alarm about rapid AI development. |
| Sept 9 | He cites a recent Hugging Face hack as heightening his concerns. |
| Sept 10 | Former OpenAI employee Pamela Mishkin urges colleagues to consider staying or leaving. |
| Sept 12 | Anthropic CEO Dario Amodei publishes a 3,800-word essay calling for prudence and a three-step safety framework. |
| Sept 11 (scheduled) | DeepMind employee Andreas Kirsch is slated to warn that AI could “kill us all.” |
Internal Momentum Across Labs
After Coxon’s departure, a dozen employees across leading labs convened on messaging platforms and private gatherings. Roughly 1,000 employees signed a petition urging stronger safeguards and a mechanism to hold companies accountable.
Official Statements & Responses
- Chief scientist Jakub Pachocki called for safety rules enforced by third-party auditors, government agencies or international bodies.
Data & Statistics
- 1,000+ employees signed a safety-focused petition (internal reports).
- 3,800-word essay by Amodei (Sept 12) outlines a three-step framework for slowing development.
- Six AI workers reported urging executives to pause development after Coxon’s resignation.
- Two companies—Anthropic and OpenAI—are preparing IPOs within the next year, creating tension between safety and market incentives.
Why It Matters
The push for a coordinated slowdown raises practical and regulatory challenges. Companies fear antitrust liability if they formally agree to limit development, yet they confront competitive pressure from Chinese start-ups releasing comparable models. Congressional interest adds political weight, while the lack of a clear monitoring mechanism leaves any pact’s shape uncertain.
Criticism & Opposition
- Roy Bahat, head of Bloomberg Beta, notes that “the sober researchers inside the labs … are now sounding terrified.”
- Rep. Ro Khanna (D-CA) calls the internal debate “a wake-up call” and argues voluntary industry measures are insufficient without legislative oversight.
Conflicting Reports & Gaps
Sources differ on the concrete form a coordinated pause would take. OpenAI’s Altman notes uncertainty about other labs’ agreement, while Anthropic’s plan to embed permanent third-party reviewers lacks detail on enforcement. No definitive timeline for a formal pact has been disclosed, and antitrust exemptions remain speculative.
Verbatim Quotes
- “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” — Dario Amodei, Anthropic CEO
- “The sober researchers inside the labs who have prided themselves on being evenhanded are now sounding terrified,” — Roy Bahat
- “It’s a wake-up call to Congress and our government,” — Rep. Ro Khanna
What’s Next
Altman hinted that a multi-company safety pact is “likely to happen,” while Amodei’s essay signals willingness to embed independent evaluators inside frontier labs. Both executives suggest any coordination will need government consent to avoid antitrust violations, moving discussions from private chats toward public policy.