Anthropic says it may soon hit the recursive self-improvement stage

Anthropic revealed in its updated S-1/A filing that it may soon reach its own standard for Recursive Self-Improvement. This is a stage where AI trains itself so well that its capabilities grow explosively.

The filing shows Anthropic already uses AI in research. It once built new models with big capability jumps without noticing right away. Based on internal estimates, its next-generation models could soon hit the "AI-driven explosive progress" line in its Responsible Scaling Policy. No current model has crossed it yet.

Anthropic stays cautious, calling this both an opportunity and a risk. It promises not to let Claude autonomously design frontier models without real human oversight, and may revise its policy to keep humans in control.