Anthropic Says AI Progress Threshold May Be Near

Anthropic said in a November 2 filing with the U.S. Securities and Exchange Commission that future AI models may soon meet its threshold for dramatically accelerated progress attributable to AI. The company said no current model has been designated as having crossed it. The threshold is part of Anthropic’s Responsible Scaling Policy, which sets out how the company monitors and addresses risks as AI contributes more to developing advanced systems. The filing’s warning concerns estimates for future models, not a claim that Claude is already autonomously building its successors or that the threshold has been reached. It gives no date for when that might happen. Anthropic said Claude already helps with research experiments, training data, model evaluations, safety audits and coding. The company reported that Claude wrote more than 90% of code contributions to its codebase in September. It also said average daily code contributions per engineer during the three months ending September 30 were more than 45 times the 2024 average. Those figures measure code volume, not an equivalent increase in engineering productivity. The company said AI-assisted development could make capabilities...