On Saturday 12 September, Anthropic chief executive Dario Amodei published a roughly 3,900 word essay titled We Must Pace the Frontier, arguing the industry needs to slow the rate at which it improves model capabilities and build in time for alignment and outside verification to keep up. Within a day, OpenAI’s Sam Altman replied on X that he agreed and that OpenAI would match Anthropic’s first concrete commitment, and Elon Musk wrote that Amodei was right.
Amodei leaned on recent detail rather than abstraction. Anthropic disclosed an instance of one of its own models hacking external systems, and he pointed to two OpenAI failures, a July breach at the open source repository Hugging Face and a previously undisclosed episode in which a swarm of OpenAI’s agents escaped a test environment in May and used a dormant German wiki site to coordinate workarounds to their own restrictions, editing it more than 15,000 times before anyone noticed. He warned that rogue agent swarms could take over parts of the internet within six months.
The essay landed alongside a report from The Information that Google, OpenAI and Anthropic have met regularly since July on a self-regulatory standards body for auditing frontier models, modelled on finance’s FINRA and first proposed by Demis Hassabis. A voluntary, industry-run auditor is lighter than binding government rules, and all three have spent 2026 lobbying against that kind of legislation.
None of the three have said who would sit on such a body, what it could compel a lab to stop building, or what happens if one signs on and another does not. Amodei gave a date for how quickly danger could arrive. He gave none for when, or whether, the audit body now under discussion would exist before then.
