Anthropic Reveals Claude Now Leads 26% of Its R&D Tasks, Up From Under 1% in February
FinanceYF5 · x · 2026-09-21
Anthropic has disclosed new data on how close it is to recursive self-improvement:
- Claude now leads 26% of R&D tasks, up from under 1% in February; it participates in or leads over 90% of tasks
- Roughly 30,000 agents continuously work on R&D
- In August, online monitoring reviewed over 1 billion decisions and blocked just 0.002%
- Offline systems flag 100,000 records weekly, with only 50 escalated to human review
The numbers offer a rare look at the scale of AI-driven internal R&D and the filtering ratios of its oversight systems.
Related event: Anthropic Says Claude Now Drives 26% of Its Internal AI R&D(3 posts)→
More from AGI Musings
- ASU's CogniShield rethinks assessment for the GenAI era — keviv9 · 2026-09-21
- After IIT Bombay tragedy, scholar calls for institutional support for faculty — keviv9 · 2026-09-21
- Universities must discuss GenAI assessment beyond detection and policing — keviv9 · 2026-09-21
- Redesigning assessment so integrity, learning and fairness coexist in GenAI era — keviv9 · 2026-09-21
- Scholar responds to IIT Bombay tragedy: time to think about solutions — keviv9 · 2026-09-21
- CTJ Lewis publishes short refutation of AI x-risk hysteria, challenging takeover logic — basedjensen · 2026-09-21