Anthropic Prospectus: AI-Accelerated R&D May Be About to Breach Its Safety Policy Threshold

November 3: According to the S-1/A filing Anthropic submitted to the U.S. SEC, the company said in its risk factors that it increasingly uses AI to accelerate its own R&D, and that model capabilities may advance faster than safety and alignment measures are developed; based on internal estimates and expectations for future models, the company may be about to reach the definition in its Responsible Scaling Policy (RSP) of "AI causing a dramatic acceleration in progress." Anthropic said it has not yet determined that any existing model exceeds that threshold, is still evaluating the relevant data, and noted that in the past it has only noticed significant R&D acceleration after it had already occurred. The company plans to keep humans in the loop in R&D and is considering updating the RSP. According to TechCrunch, the statement is seen as a warning that Anthropic is approaching recursive self-improvement (RSI) capabilities.