Anthropic has said its AI model Claude now leads 26 per cent of the company’s AI research and development work, up sharply from less than 1 per cent in February, in a sign of how quickly AI systems are taking over the process of building future AI.
The company disclosed the figures through a new measurement tool called the Anthropic R&D Automation Index, which tracks how much of its AI research is performed by Claude versus humans. The index uses an automation scale developed by Epoch AI, ranging from AL0, where AI has no involvement, to AL5, where AI operates fully autonomously with no human in the loop.
At the AL3 “collaborates” level, AI can complete large portions of a task under close human direction. At AL4, the “leads” level, AI can complete most of a task end-to-end from a high-level prompt while a human supervises. As of August, Claude had reached the AL4 level for 26 per cent of Anthropic’s AI R&D work, while more than 90 per cent of all R&D work was being performed at or above the “collaborates” level. Claude has not yet reached full autonomy in any measured area.
Anthropic also disclosed that around 30,000 AI agents were carrying out research and engineering work on its internal platform at any given time as of August. The company said all of these agents’ actions pass through an online safety monitor before execution, and are also reviewed afterward by a separate offline system. Of more than 1 billion agent decisions analysed in August, roughly one in 47,000 was blocked by the online monitor, while about 100,000 transcripts are flagged weekly for further review, with roughly 50 high-priority cases escalated to humans each week.
The company said around 6 per cent of computing power used for AI R&D was allocated to safety work during a sample week in July, rising to about 12 per cent for AI-driven R&D specifically.
Anthropic said it plans to keep publishing these measurements and bring in independent evaluators to verify safety practices, as the industry edges closer to AI systems capable of improving themselves with less human oversight.
(Source: Business Standard)
