Artificial intelligence safety company Anthropic announced that its Claude model now autonomously executes more than a quarter of the firm's internal research and development tasks.
According to a newly published measurement index, Claude independently carries out twenty-six percent of model development tasks from high-level prompts while operating under human supervision. The company also disclosed that Claude assists in approximately ninety percent of its total research and engineering workflow. Anthropic emphasized that the system does not yet possess full recursive self-improvement capabilities because human oversight remains required throughout the training pipeline.
The disclosure comes as Anthropic chief executive Dario Amodei continues to advocate for industry-wide transparency and mandatory reporting standards regarding advanced model progress. Experts warn that tracking the exact degree to which AI models design and refine their successors is vital to preventing uncontrolled runaway capability leaps.