Is Big Tech’s AI slowdown a safety pact or a cartel?

After Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk loosely agreed to slow AI development, critics asked whether “pace the frontier” was safety leadership or a way to freeze competitors and dodge harder law. Amodei’s essay floated third-party auditors, domestic lab rules, and a global slowdown agreement.

Safety researchers told The Verge the direction is overdue — many had pushed for it for years, including after the OpenAI–Hugging Face agent incident — but nearly all said verbal nods are worthless without enforceable limits. Under Trump, heavy federal AI regulation looks unlikely.

We’re with the camp that calls this promising only if compute, audits, and RSI speed limits become contracts, not vibes.

We’re not joining the pure-cynicism chorus, and we’re not buying CEO sainthood. Ex-OpenAI researcher Daniel Kokotajlo’s read matches what we’ve seen: outsiders yelled for a slowdown for years; CEOs are now absorbing the pressure and claiming the mic. Apollo’s Marius Hobbhahn called it one of the best safety ideas in a long time if it actually happens. That “if” is the whole story.

Safety-washing risk is real. Labs could invite auditors, generate paperwork, and barely change release cadence. Social platforms ran a similar call-for-rules play to shape softer outcomes. Meanwhile Trump labeled AI fears a hoax on speakerphone with Jensen Huang — a reminder that voluntary frameworks may be the only game in town for now.

China remains the rhetorical trump card against any pause. Chinese officials called slowdown talk fearmongering. Useful international coordination would still be historic if it happened; waiting for Beijing is also a convenient excuse to change nothing. Recursive self-improvement is the underlying scare — Amodei says RSI is starting across the industry and needs a speed limit. I’m watching for auditor embed rights with whistleblower paths, compute-budget cuts you can measure, and whether open-source is carve-outed rather than kneecapped.

Anthropic researcher Jacob Coxon’s resignation letter — warning labs are racing to self-improving superintelligence — drew enormous attention, including more than 170 million views on X. A July letter from more than 1,000 lab employees also called for a slowdown after Hugging Face. Amodei proposed embedding auditors such as METR, Apollo, and Redwood.