Anthropic reports Claude contributes to over a quarter of its AI R&D
Anthropic said its AI system Claude accounted for 26% of internal research and development work, while human-AI collaboration covered more than 90% of projects in August.
Anthropic disclosed that its AI model Claude was responsible for 26% of the company’s artificial-intelligence research and development effort, according to a blog post released on Thursday. The firm said Claude’s role has grown from 1% in March, based on a scale created by the nonprofit Epoch AI that monitors AI progress.
Human-AI collaboration accounted for over 90% of the research activity at the San Francisco-based startup in August. About 30,000 AI agents were active on Anthropic’s internal platform at any given time, and each proposed action was screened before execution. Of more than a billion decisions made that month, roughly one in 47,000 was blocked.
Anthropic also reported that 6% of its computing power used for AI research in a sample week in July was allocated to safety work, rising to 12% for research conducted by AI itself. The company noted that the figures are conservative because computing used for both safety and capability was counted as capability work.
The disclosures come as industry peers, including OpenAI, have begun publishing reports on unexpected model behavior, highlighting growing scrutiny of AI autonomy and safety.