Anthropic Says Its Recursive Self-Improvement Alarm Has Been Triggered
Reported by Gizmodo1 source ↓
Anthropic says it has met a threshold in its Responsible Scaling Policy for AI-accelerated research progress, based on what it expects from future models—not on progress observed in current models. The filing does not say recursive self-improvement is happening.
Key points
- Anthropic says its AI systems have already accelerated its research and development
- The company says its threshold assessment rests on expected future progress, not observed progress from current models
- Crossing the threshold calls for mitigations under Anthropic’s Responsible Scaling Policy
Details
In a November 10 version of its SEC filing, Anthropic said its internal estimates indicate that AI-driven research progress may increase further. The company said it determined that it met the relevant threshold because its policy applies when it observes or expects a specified degree of accelerated progress.
The filing’s November 2 version said Anthropic may imminently meet criteria for “dramatically accelerated progress attributable to AI”; the November 10 version removed that phrase. Gizmodo reports that the policy threshold concerns a doubling of the baseline rate of improvement, and that crossing it calls for measures including stronger security, monitoring model activity and improving the company’s ability to red-team its own systems.
Written from coverage as of
Images
- Gizmodo
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh