Anthropic says in its IPO filing that it has reached the “rapid acceleration” threshold for doubling the pace of AI development through research automation, based on expectations for future models and its overall pace of progress rather than observations of current models
Anthropic says in its IPO filing that it has reached the “rapid acceleration” threshold for doubling the pace of AI development through research automation, based on expectations for future models and its overall pace of progress rather than observations of current models
On November 10, Anthropic filed Amendment No. 2 to its IPO filing with the US Securities and Exchange Commission. In Amendment No. 1, dated November 2, it suggested that it might soon reach the threshold, which its August risk report had assessed as still unmet.
Anthropic’s Responsible Scaling Policy sets out in advance which safeguards take effect as model capabilities increase. The “rapid acceleration” threshold is reached when the company observes or expects AI progress to proceed at twice both the expected rate and the fastest previous rate achieved without substantial AI assistance. Amendment No. 2 applies that definition as follows:
“We have determined that we have reached this threshold based on what we expect from future models and our overall pace of progress, rather than what we have observed in current models.”
In the same amendment, the company acknowledges that much of its evidence on how much AI accelerates development arrives with a delay. It has previously recognized substantial acceleration only in retrospect and may miss it again. The paragraph announcing the decision drops the statement in Amendment No. 1 that the company had not identified any current model as having crossed the threshold. It adds that “our use of AI systems is already accelerating our research and development.”
On November 5, METR, an independent organization studying AI risks, estimated that Anthropic’s aggregate benchmark performance has been improving at 1,9 times its previous rate since October 2025. METR estimated that AI accounts for about 80% of the current acceleration. Anthropic redacted most of the evidence METR found most informative from the public assessment. METR said this evidence led it to take roughly twofold acceleration more seriously and consider the possibility that acceleration exceeds that level and is increasing. METR could not disclose its forecasts of future model performance.
On November 9, the day before the filing, Anthropic released a technical report on Claude Mythos 5.2. For the same metric, it reports an acceleration factor of 1,88, formally below the threshold. The company does not consider the model to have reached the threshold. According to the report, reaching the threshold also depends on “expected future progress.”
The filing describes the cycle in which each version of Claude helps build the next in two ways. The business section calls it “one of the most significant structural advantages.” The risk section describes a danger that could cause the company to “lose visibility into individual processes in ways that make them difficult for any person or group of people to understand and control.” The company is considering a policy update that would require human approval for certain research and development decisions.
The policy ties the threshold to safeguards, including “eyes on everything”: AI systems review logs of all critical actions in AI development for security threats and concerning behavior by people and AI systems themselves. According to its safety roadmap, the company expects to approach that level of oversight by January 1, 2027. By January 15, it pledges either to publish a proposal for giving governments direct influence over the pace of AI development or to explain why it has chosen not to do so.
In their letter to investors, the seven cofounders acknowledge that they may slow development, particularly when its pace is set by systems that build themselves: “There may be times when we choose to move more slowly than we could to make our systems safer, even if doing so puts us at a competitive disadvantage.”
- https://www.sec.gov/Archives/edgar/data/2130357/000162828026072484/anthropicpbc-sx1ax2.htm
- https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm
- https://t.me/UkhvatNews/4823
- https://assets.claude.com/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
- https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/
- https://t.me/UkhvatNews/4889
- https://assets.claude.com/27d60918d218125e7504bc7354c4b4626ec56b11.pdf
- https://www.anthropic.com/responsible-scaling-policy/roadmap
- https://ipo.anthropic.com/documents/eu/founders-letter.pdf