Anthropic Warns It May Soon Reach Its AI Self-Improvement Threshold

Originally published on The Daily Flare.

Anthropic warned in an updated US securities filing on Monday, November 2, that it may imminently reach its own threshold for AI-driven acceleration in developing new models.

The warning appears in a revised S-1/A filing with the US Securities and Exchange Commission. Anthropic said its internal estimates and expectations for future models, rather than its current systems, suggest it may soon meet the Responsible Scaling Policy’s definition of “dramatically accelerated progress attributable to AI.” The company said it has not designated any current model as exceeding that threshold and is still assessing the underlying data.

The concern is linked to recursive self-improvement, or RSI: a process in which AI systems help develop more capable systems, potentially accelerating progress further. Anthropic’s filing describes AI-assisted research and development as an area covered by its safety policy. TechCrunch reported that the policy’s threshold refers to a dramatic acceleration in progress attributable to AI, rather than simply an AI system helping with ordinary research tasks.

Gizmodo’s account of the policy says the benchmark is a doubling of the baseline rate of improvement, with that increase plausibly attributable to AI. Anthropic uses the phrase “dramatically accelerated progress attributable to AI” in its filing, while identifying recursive self-improvement as the broader risk it is monitoring. The filing frames the possibility as a forecast about future systems, not a claim that the threshold has already been crossed.

Anthropic also disclosed that its growing use of AI in research could let model capabilities advance faster than the company can develop, validate and implement safety, alignment and security measures. It warned that it has previously failed to recognize a significant acceleration in development until after it occurred, and may fail to identify one in the future until then.

The company said it has already used AI to help develop new models whose capability gains it did not recognize at the time. That disclosure gives the warning a present-day basis, even as Anthropic says no current model has been designated as reaching its RSI threshold.

Anthropic’s filing states: “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress.” The company said its Responsible Scaling Policy sets out plans to monitor and mitigate risks as it approaches what is known as recursive self-improvement.

India Today reported that, as of May 2026, Claude was writing more than 80 per cent of the code used in Anthropic’s own systems. The outlet said human engineers were reviewing and supervising the AI’s work. That reported use of Claude in the company’s systems is separate from the filing’s assessment that future models may soon meet the more specific acceleration threshold.

Anthropic said it does not expect to be able to safely use Claude to “fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible.” It said it is considering changes to its Responsible Scaling Policy intended to keep humans in control of frontier-model development.

The warning reflects both the potential gains and risks associated with AI-assisted research. TechCrunch described RSI as a possible source of substantial productivity and scientific discovery, while also noting concerns that rapid capability growth could outpace human control. The filing itself focuses on Anthropic’s assessment of its own development pace and the safety measures needed as AI takes on a larger role in research.

Risk disclosures in securities filings are intended to alert investors to potential hazards. TechCrunch noted that claims about approaching a powerful capability can also be viewed skeptically because such filings may serve a promotional purpose as well as a disclosure function. Anthropic’s filing nevertheless states that its current models have not crossed the company’s threshold.