Anthropic signals proximity to self-improving AI capability in new filing

Anthropic disclosed in its revised S-1A filing that the company is nearing a point of recursive self-improvement (RSI), a threshold where AI accelerates its own development dramatically. While the filing detailed soaring revenue alongside rising expenses, the core warning centered on the risks associated with AI assisting in building more capable AI systems. The company noted that its increasing use of AI in research and development could cause model capabilities to advance faster than the ability to implement corresponding safety and alignment measures. Anthropic stated that internal estimates suggest they may soon meet their Responsible Scaling Policy’s definition of dramatically accelerated progress attributable to AI, though they confirmed no current models have crossed this threshold. Despite these cautions, Anthropic indicated that it does not foresee being able to safely design and build successive frontier models autonomously without meaningful human oversight. This suggests the company is actively considering updates to its Responsible Scaling Policy to maintain human control over frontier model development.