Anthropic told TechCrunch on Monday that its revised S-1A now includes new Risk Factors tied to recursive self-improvement, a point where AI could begin accelerating its own development. The filing says future models may imminently meet its RSI threshold, even though current models have not crossed it yet.
The company says its increasing use of AI systems in R&D could push model progress faster than its ability to build matching safety, alignment, and security measures. For readers watching frontier models, that means the gap between capability growth and control mechanisms is now part of Anthropic’s own disclosure, not just outside speculation.
Anthropic and RSI
The filing says Anthropic may imminently meet its Responsible Scaling Policy definition of “dramatically accelerated progress attributable to AI.”
In plain terms, the company is saying its internal estimates for future models suggest a threshold may be close enough to matter operationally, not just theoretically.
Claude under supervision
Anthropic said it does not anticipate being 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.
That sentence gives the clearest line in the filing: even if AI becomes capable of taking over more of the research loop, Anthropic is telling investors that human oversight stays necessary for as long as the company can enforce it.
The filing also says Anthropic is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.
That is the practical next step for readers tracking safety governance, because the policy is the mechanism Anthropic uses to decide when to slow down, add safeguards, or change how advanced systems are handled.
Anthropic also said it has not designated any of its current models as exceeding the RSI threshold yet, while continuing to assess the underlying data.
That leaves the company in a narrow middle ground: it is warning that future models may soon reach the line, but it is not saying today’s models have already crossed it.
The unresolved issue is what specific evidence led Anthropic to conclude that its future models may imminently meet its RSI definition, and the filing does not say when any policy updates will be made.
