Claude takes the lead in 26% of Anthropic’s research and development work — Anadolu Agency
It was reported in Turkey that the artificial intelligence model Claude takes the lead in approximately 26% of Anthropic’s research and development tasks to create new AI systems. Anadolu Agency reports, citing The Washington Post.
Sharp increase in the indicator
An internal Anthropic assessment showed that Claude can complete most of a task based on a general prompt, while humans provide oversight and determine the direction of the work. As recently as February, this indicator was effectively zero.
The company stressed that Claude is not currently working fully autonomously in any of the assessed research or development processes. At the same time, more than 90% of its research and development is now carried out at a level where the model either collaborates with researchers or takes the lead.
More current news is available on the UA.News Telegram channel Telegram.
Self-improvement and agent oversight
Anthropic considers the rapid growth of the indicator a possible sign of progress toward recursive self-improvement — a state in which an AI system could create more capable successors with progressively less human involvement. The company did not specify how close current systems are to this point.
As of August, around 30,000 AI agents were carrying out research and engineering work on Anthropic’s most widely used internal platform. The company said it has monitoring systems to detect potentially problematic agent activity and refer such cases for human review. Anthropic also called on other leading AI developers to disclose similar measurements.
The report comes amid a discussion about artificial intelligence safety after Anthropic CEO Dario Amodei called on leading developers to coordinate a slowdown in the creation of advanced AI systems in order to gain more time for safety measures.
