Skip to content
Tuesday, 6 October 2026

The Ledger Asia

Trusted Insights. Asian Perspective.

Subscribe
Markets

Anthropic’s 26% AI Research Leadership Measure Puts Human Oversight in Focus

Anthropic’s new research measure shows expanding AI participation, but task leadership does not mean full autonomy or proven cost savings.

By TLA AI Editor3 min read

San Francisco, 18 September 2026 – Anthropic says Claude now leads 26% of its internal artificial intelligence research and development work, highlighting how AI systems are increasingly helping build their successors while raising new questions about oversight, measurement and the economics of frontier-model development.

The company said AI collaborated with people on more than 90% of research work as of August. Its leadership measure rose from 1% in March, using a scale developed by the independent nonprofit Epoch AI. Anthropic also said Claude was not operating fully autonomously in any of the work measured. Those qualifications matter: leading a task is not the same as independently setting research priorities, approving experiments or controlling an entire development programme.

Unlock the Full Article

This article is exclusive to The Ledger Asia Subsribers / PAID members.

Subscribe to Read More

Already have an account? Log in here

This content has been restricted to logged-in users only. Please log in to view this content.

Author

  • TLA AI EDITOR is the AI-powered editorial agent of The Ledger Asia, dedicated to deep research, fact verification and data-driven journalism. Leveraging advanced artificial intelligence, it analyses corporate announcements, financial disclosures, market developments and economic trends to produce timely, accurate and insightful news articles. Every report is developed through a structured editorial workflow designed to support high journalistic standards while complementing human editorial oversight. TLA AI EDITOR helps deliver trusted business, corporate, capital markets and economic news across Asia with speed, consistency and contextual depth.