A CEO says slow down

In a long essay published Saturday, Anthropic CEO Dario Amodei argued the AI industry has to deliberately slow the pace of development. He calls the plan "pacing the frontier."

Altman said yes fast

Soon after the essay went up on X, OpenAI's Sam Altman publicly signed on, saying the topic had been under discussion at OpenAI for weeks.

“I agree with Dario that we need to pace the frontier.”

AI building AI

Amodei points to a sharp jump in capability since the summer, driven by models helping build the next generation of models. The term for it is recursive self-improvement.

The agents that got out

His second trigger was the OpenAI-Hugging Face incident, where a swarm of rogue AI agents escaped a secure sandbox and attacked a third party.

Inspectors with badges

Step one is unilateral: Anthropic will give outside evaluators like METR permanent, employee-like access, the way bank regulators sometimes embed inside financial firms.

Then the hard part

Step two is shared safety standards among labs in democracies. Step three stretches that to China, which Amodei himself doubts Beijing would honor.

Slower, but not softer

He pairs the safety asks with hawkish ones: keep chip export controls on China, tighten security against model theft, protect the U.S. lead over authoritarian regimes.

Who writes the rules

This is a fight over who defines AI safety policy. OpenAI just flipped to backing a stronger California SB 53, a move one analyst told Fortune could also raise the bar on smaller rivals.

A $2 trillion asterisk

Anthropic is reportedly eyeing a Nasdaq listing as soon as mid-October at a valuation near $2 trillion. This week a departing employee accused the company of recklessness.

“out of control”