Anthropic reports 26% of R&D reaches AI-led stage with 30,000 agents
Anthropic announced on September 17 that 26% of its research and development work now falls under AL4—where AI handles most tasks independently, with humans providing only broad instructions. This marks a 26-fold increase from less than 1% in February 2026, according to the company’s own AL0–AL5 scale, developed by the US research institute Epoch AI. Over 90% of Anthropic’s R&D now involves AI…
Key points
- Anthropic says 26% of R&D now reaches AL4, where AI leads with minimal human input, up from <1% in February 2026
- 30,000 agents run continuously, monitoring 1B+ decisions with 0.002% blocked and 50 human reviews per week
- 6% of compute resources were used for safety research in a July survey, per company data
The company operates 30,000 agents continuously, monitoring over 1 billion decisions with a 0.002% block rate (about 2 blocked actions per 47,000). Humans review 50 high-priority cases weekly, while safety research consumes 6% of computing resources, per a July survey. The author tested the same scale on their own 11-step article creation process, finding 90% at AL4, but clarifies this doesn’t imply superiority—repetitive tasks like drafting are easier to automate than open-ended research.
[Verification] 26% of Anthropic's R&D was led by AI—when I measured my own daily posts with the same yardstick, it was 90%
note.com · 27 September 2026
Loading the full article…
This text was published by note.com and written by やすだ.dev@毎日投稿. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
3sourcesThe headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Elorian and Scale AI release benchmark showing AI lags in visual reasoning · 1 src
- AI labs buy data from dead startups for training models · 1 src
- Hugging Face’s ML-intern builds custom models from prompts in days · 1 src
- Coding Agents Can Solve the Task — But How Much Work Did They Waste? · 1 src
- Anthropic research suggests small AI models may self-replicate on IoT devices · 5 src
Comments
via GitHub Discussions