Anthropic tells investors it may 'imminently' meet its own threshold for self-accelerating AI
New risk-factor text says no current model has crossed the line, but admits past capability jumps were only spotted after the fact.
The amended S-1 filed Monday adds language to its risk factors that was not in the October 26 version, TechCrunch reported. Anthropic says its growing use of AI to speed up its own research could make its models advance faster than its safety, alignment and security work can keep up, and that it 'in the past' failed to notice significant accelerations until after they had happened. It then says its internal estimates and expectations for future models, not current ones, indicate it 'may imminently meet' its Responsible Scaling Policy's definition of 'dramatically accelerated progress attributable to AI,' the threshold the policy ties to recursive self-improvement.
Anthropic stresses that it has not designated any current model as exceeding the threshold and is still assessing the data. Under the policy, the bar is roughly a doubling of the normal rate of AI progress, a year's work producing what would usually take two, that can plausibly be credited to automated research. The filing also says Anthropic does not expect to be able to safely let Claude design and build later generations of frontier models fully on its own, without meaningful human supervision, and that it is considering changes to the policy to keep humans in control.
As TechCrunch notes, a prospectus is both a legal disclosure and a sales document, and the underlying data cannot be checked from outside the company. Still, it is rare for a frontier lab to tell securities regulators, under liability for misstatements, that it may be close to a risk line its own safety policy defines.
- Filing
- S-1/A, Nov 2, 2026
- Threshold
- RSP 'dramatically accelerated progress attributable to AI'
- Current models over the line
- None designated
Read the full story at TechCrunch
Sources TechCrunch