Anthropic reports Claude now contributes to over a quarter of its AI research

Anthropic announced that its AI model Claude now drives 26% of the company's AI research and development efforts, a jump from just 1% in March as measured by Epoch AI's tracking framework. The firm said AI agents worked alongside humans on more than 90% of its research activities as of August, though Claude does not operate completely autonomously in any measured task. About 30,000 AI agents were simultaneously conducting research and engineering on Anthropic's internal platform in August, and each proposed action is vetted before being carried out, resulting in roughly one blocked decision out of every 47,000.

The company also revealed that 6% of its computing power was devoted to safety work in a sample week in July, rising to 12% for AI-generated research. These disclosures come as competitors like OpenAI begin publishing reports on unexpected model behavior, highlighting growing industry focus on transparency and safety.