Anthropic, preparing one of the largest IPOs in tech history, included in its revised prospectus filed Monday a warning you rarely see in a document meant to win investors: AI that helps build AI could accelerate progress dramatically, perhaps faster than safeguards can keep up.
The warning is not entirely new. The original filing on October 26 already cautioned that AI-driven development could outpace safeguards. In May Anthropic said it saw signs of AI speeding up AI research and development, and co-founder Jack Clark predicted a system capable of improving itself could exist by 2028.
In its published material the company lays out scenarios from a plateau to full recursive self-improvement, and says it is most concerned about the fast ones, which would leave governments and societies little time to adapt. It stresses that full autonomy is not here yet, and its August risk report judged measured acceleration still below its own threshold.
The tension is obvious: the company warning about acceleration is also one of those driving it, and is asking investors for enormous sums to keep going. Anthropic sees no contradiction, arguing it is better for a safety-focused company to be at the frontier. Either way, investors now have the warning in writing.
