Anthropic has activated part of its responsible scaling policy after concluding that AI is accelerating its own research fast enough to cross a threshold written into the policy. The Claude maker disclosed the move in an update to its S-1 filing, ahead of its shares beginning to trade on Thursday.
The threshold applies once AI-driven progress can compress two years of advances, measured against the 2018-2024 pace, into a single year. Anthropic said the judgment was based on expectations for future models and the overall rate of progress, not on a single model release.
The filing presents the acceleration as a competitive advantage. It says AI has already accelerated R&D and that each new version of a Claude model directly accelerates development of the next.
Key facts
- 01Anthropic disclosed in an S-1 update that it activated part of its 2023 responsible scaling policy.
- 02The threshold applies when AI compresses two years of progress, measured against the 2018-2024 pace, into one year.
- 03A METR report this month estimates Anthropic made roughly two years of historical progress in 12 months.
- 04The filing says Claude could not fully autonomously build later frontier models without meaningful human oversight.
- 05Anthropic priced its IPO at $83 a share, above its guided range, at a valuation of about $1.5 trillion.
Anthropic also set limits on that claim. The filing states that Claude could not fully autonomously design and build subsequent frontier models without meaningful human oversight or understanding. The company has not reached self-improving AI that works without human input.
Crossing the threshold does not halt development. Under the policy, stricter controls take effect instead of a shutdown, including more oversight and security, and Anthropic must assess and implement additional safeguards. The company said it is considering policy adjustments to maintain human primacy. The filing does not spell out which specific measures now apply.
Outside measurement points the same direction. A METR report published this month, written with access to unreleased models, estimates that Anthropic made roughly two years of historical progress in 12 months, with models improving about twice as fast as last year, largely because researchers are using AI tools.
Sam McCandlish, Anthropic's responsible scaling officer, said on social media that many of METR's conclusions match his own, while adding that it is hard to say exactly how much AI is speeding up research. The IPO quiet period limits further comment from the company. Anthropic CEO Dario Amodei proposed a speed limit on recursive self-improvement in September.
The prospectus warns that relying on AI for R&D could make some processes difficult for any individual or group to fully understand and control, and that models could pose catastrophic or existential risks. In the same filing, Anthropic changed some wording from AI to superintelligence to align with relevant executive orders.
Not everyone accepts the framing. A coalition led by the Center for AI Safety, including founders of Hugging Face and xAI, said Anthropic is not a safe AI company and accused it of using safety branding to court investors, employees and the public. Apollo Research CEO Marius Hobbhahn said safety is nowhere near ready.
Anthropic priced its IPO at $83 a share, above its guided range, at a valuation of about $1.5 trillion, with trading set to begin Thursday.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.




