Claude now leads 26% of Anthropic’s AI research, development work: Report

Anthropic's artificial intelligence (AI) model Claude is now leading more than a quarter of the company's research and development work on new AI systems, The Washington Post reported Thursday.

An internal review found that Claude "leads" about 26% of Anthropic's AI research and development, meaning it can complete most of a task from a high-level prompt while humans provide supervision and direction, according to the report. The figure was effectively zero as recently as February.

Anthropic said Claude is not currently operating fully autonomously in any of the AI research and development work it measured. However, more than 90% of the company's R&D is now conducted at a level where Claude either collaborates with researchers or takes the lead.

The company said the rapid increase could provide an indication of progress toward "recursive self-improvement," in which an AI system becomes capable of developing more advanced successors with increasingly limited human involvement.

Anthropic did not say how close it believes current systems are to reaching that point.

The company also disclosed that about 30,000 AI agents were conducting research and engineering work on its most-used internal platform as of August. Anthropic said it has monitoring systems designed to detect and flag potentially problematic agent activity for human review.

It has called on other leading AI developers to publish similar measurements, arguing that greater disclosure would allow governments, researchers and the public to better track the pace of AI development.

The disclosure comes amid renewed debate over AI safety after Anthropic CEO Dario Amodei called for leading developers to coordinate a slowdown in frontier AI development to allow more time for safety measures.