Anthropic’s amended S-1/A, which Clubic reported was dated November 2, 2026, warned that automating AI research and engineering could accelerate model progress. Clubic described a prospective threshold in Anthropic’s Responsible Scaling Policy: a doubling in the pace of progress, equivalent to one year of advances that would otherwise take two. The report said Anthropic had not declared a current model beyond that threshold.
What an automated research loop would involve
A full loop would cover choosing experiments, writing code, training models and analyzing results. Clubic reported that human researchers’ judgment remains important in deciding which research directions to pursue.
Clubic also described risks the framework considers, including errors or biases carrying over to later models and becoming harder to detect, as well as potential manipulation of successor-model training or safety evaluations.
Why code output is not the same as self-improvement
ABC reported that Claude generated more than 90% of Anthropic’s code during September. The report also put average daily code output per engineer at more than 45 times the prior-year level. Those figures describe code output; the policy threshold concerns the rate of model progress.
For context on Claude’s reported role in coding, NeoTeo previously covered its use by Anthropic’s platform engineering team. The team-specific account and ABC’s company-wide figure describe different scopes.
A June report by The Verge described recursive self-improvement as an AI system fully autonomously designing and developing its own successor. Code generation, even at a large reported scale, is not the same as automating that full process.
Reddit commenters questioned the warning’s timing and possible IPO-related messaging; others argued that potential commercial incentives were no reason to dismiss the technical risk.
