Breaking

Anthropic Says Claude Now Leads 26% of Its AI R&D Work Under Human Supervision

September 17, 2026 at 22:25 EDT

Foundation Models
AI Agents
Policy & Regulation

Anthropic said Claude now leads 26% of its measured AI research and development work, up from about 1% in March, while still operating under human supervision. The disclosure is part of a new transparency program meant to show how much frontier AI labs are using AI to build more capable AI systems.

Sep 2026 · Anthropic

Claude is now part of the AI lab

Anthropic says Claude leads 26% of its measured AI R&D work while humans still supervise. The disclosure turns internal AI development into a set of public metrics.

26%

AI R&D Claude now leads

>90%

Work where AI collaborates or leads

30,000

Internal agents at a time

1:47k

Online monitor block rate

Lead share moved from 1% to 26%

Same unit: share of measured AI R&D work led by Claude.

What Anthropic is trying to measure

1Map work

Create a task tree for model R&D.

→

2Rate automation

Classify whether AI assists, collaborates, leads or acts alone.

→

3Watch agents

Monitor actions before and after execution.

→

4Track compute

Estimate the safety share of AI R&D resources.

Why it matters

The figures turn rapid AI-agent adoption inside a frontier lab into measurable operational data.

What remains limited

No measured R&D category is fully autonomous, and the method still needs common standards and outside verification.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.