Anthropic Says It May Be Near Its Definition of Recursive Self-Improvement

Anthropic says it may soon meet its internal definition of progress accelerated by artificial intelligence, while emphasizing that none of its current models has crossed that threshold.

In a revised S-1 filing, the company said increasing use of AI in research and development could make model capabilities advance faster than its ability to develop and apply safety, alignment and security measures. It also said it has previously failed to identify significant acceleration in development until after it happened.

Anthropic’s internal estimates, based on expectations for future models rather than current ones, indicate it may “imminently” meet its Responsible Scaling Policy’s definition of “dramatically accelerated progress attributable to AI.” The company said it has not designated any current model as exceeding the threshold and is still assessing the underlying data.

The filing describes AI-assisted AI development as a considerable safety risk if it sharply speeds progress. Anthropic said it does not expect to safely use Claude to autonomously design and build successive generations of frontier models without meaningful human supervision or understanding. It is considering policy updates intended to keep humans in control of frontier model development.

The filing also serves as a disclosure and may be viewed as marketing, so its claims warrant context. Anthropic’s statements describe a possibility it anticipates, not a capability it says its current models have reached.