Anthropic has determined that it has met the AI research and development capability threshold in its Responsible Scaling Policy, the framework that ties its safety measures to model capabilities. The disclosure appears in the risk factors of an amended registration statement filed on November 10, the same document that set a price range of $70 to $78 a share for its initial public offering.
The filing says the threshold applies when the company observes or expects a specified degree of accelerated progress. Anthropic’s determination is based “on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models,” it says. The company adds that its use of AI systems “has already been accelerating our research and development.”
Under the policy, that means double the rate of progress in AI capabilities expected without significant AI contributions to research, where it is plausible the doubling is substantially due to automated research or engineering. Anthropic’s August risk report called such a doubling a potential early warning rather than evidence that the threat had materialized.
The previous version of the filing, from November 2, said internal estimates and expectations about future models indicated that Anthropic “may imminently meet” the policy’s definition of “dramatically accelerated progress attributable to AI.” It also said: “We have not designated any of our current models as exceeding this threshold.” Neither statement appears in the new amendment.
Anthropic says it continues to assess the underlying data, is evaluating the implications of meeting the threshold and will release further reports. The same risk factor says Anthropic’s precautionary measures have included delaying model releases and limiting their initial availability, as with Mythos through Project Glasswing. Future measures could affect “whether, how, and when” it develops or deploys new models, the filing says. Elsewhere, the filing says the company is considering updates to the policy “to keep humans in control of frontier model development.”
The disclosure follows two reports published on November 5 by METR, an AI evaluation nonprofit, after an 11-week, high-access review at Anthropic. METR said that as of October 2026, Anthropic was likely improving its models’ capabilities on well-specified tasks at around twice the rate it did between March 2024 and October 2025. It gave that estimate with high uncertainty, saying it would not be surprised if the true figure were closer to 1.3 or 3 times the historical rate. METR said its assessment did not pay special attention to the policy’s thresholds.
METR guessed that a majority of the current acceleration, perhaps about 80 percent, was due to AI, a judgment it said carried substantial uncertainty. It found that benchmark scores on Anthropic’s version of the Epoch Capabilities Index have grown at an estimated 1.9 times the earlier rate since October 2025, though it said the trend could reflect overfitting to the benchmarks. METR said Anthropic redacted a majority of the evidence it found most informative; Anthropic described the redactions as “aimed at removing confidential information that could accelerate other AI developers.”
In a separate review of the August risk report’s section on automated research, METR agreed with Anthropic that catastrophic risks from its models as of July 15 were low. It said the report’s claim that the doubling threshold had not been crossed rested on “contestable judgment calls,” and that the public report understated the acceleration under way. METR’s researchers, like the relevant Anthropic staff, leaned toward the threshold not having been crossed by July 15, but METR said it was plausible that it had been.
Sources: SEC filing, METR assessment, METR review
–
By the Control Plane Editorial Team