Anthropic's Claude is now leading 26% of the artificial intelligence research and development work at the company, up from less than 1% in February, the company said in a blog post on Thursday.

It also said that more than 90% of Anthropic's AI R&D work is now performed at a level where AI either collaborates with humans or leads the work. The company added that Claude is not yet operating fully autonomously in any measured area of its R&D.

The information is part of a new set of metrics Anthropic plans to publish regularly to give the public, governments and outside researchers greater visibility into the pace of AI development inside frontier laboratories.

The index uses an automation scale developed by Epoch AI that ranges from AL0, where there is no AI involvement, to AL5, where AI operates fully autonomously without a human in the loop.

At AL3, AI collaborates with humans and can perform large portions of a task under close human direction. At AL4, or the "leads" level, AI can complete most of a task from a high-level prompt while a human supervises.

Anthropic said Claude has reached the AL4 "leads" level for 26% of its AI R&D work as of August.

The company built its measurement by examining internal work records, including Slack and other documentation.

At AL5, an AI system would be able to identify a problem, determine what work needed to be done, develop and test a solution and deploy it without human involvement.

Anthropic said it has not reached that level in any of the AI R&D work measured.

The company added that the AI increasingly helping to develop more capable AI could accelerate progress, but could also make it harder for humans to understand or control increasingly capable systems.

About 30,000 AI agents were carrying out research and engineering work at any one time on Anthropic's most-used internal platform in August.

Anthropic said 100% of the agents' actions pass through an online monitor before execution, usually within seconds. More than 1 billion agent decisions were analysed during August, of which about 0.002%, or one in 47,000, were blocked.

A separate offline monitoring system analyses agent activity after actions have taken place. Anthropic said the system flags roughly 100,000 transcripts a week for additional automated review, with about 50 of the highest-priority cases per week escalated to humans.

The company said the monitoring system is designed to catch both immediate threats and behaviours that may develop more gradually across multiple agents.

During a sample week from July 13 to July 20, about 6% of the computing power used for AI R&D was allocated to safety work. For AI-driven AI R&D, the share was about 12%.