Anthropic warns it may imminently meet its own recursive self-improvement threshold

Anthropic's warning sits inside a regulatory filing that is primarily about money: the revised S-1A on Monday showed revenue soaring alongside rising expenses. But the Risk Factors section carries a second, separate message. Anthropic says it has already used AI to develop new models whose leaps in capability it did not recognize at the time, and it warns that its increasing use of AI in research may let model capabilities outrun its ability to develop, validate and implement safety measures. The company adds that it may not identify a significant acceleration in its development rate until after it has occurred.

The specific trigger it names is internal: its own Responsible Scaling Policy, which defines a threshold of 'dramatically accelerated progress attributable to AI.' Anthropic says its internal estimates and expectations regarding future models — rather than its current models — indicate it may imminently meet that definition, while confirming that no current model has been designated as exceeding it. It also says it is considering updates to the policy to keep humans in control of frontier model development, and that it does not anticipate safely using Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision.

The timing matters because RSI was widely treated as a buzzword or an unreachable goal when the term became popular earlier this year, and because the disclosure comes in a document that serves as both disclosure and marketing. The body notes that skeptics might read a claim of being close to a self-building model that way. Still, the filing's own framing is that the risk is real enough to be named in the fine print.