15 hrs ago
Anthropic Reports Claude Leads 26% of AI R&D Work
Anthropic is measuring how much AI helps people build new AI systems.
It says Claude led 26% of the measured work by August 2026.
That number was below 1% in February.
More than 90% of the work now includes AI in some collaborative way.
However, Claude does not yet do any measured part of the job completely by itself.
Anthropic also says about 30,000 AI agents were working at the same time on an internal platform.
Their actions were checked by monitoring systems.
The company used 6% of its total AI research computing power for safety during one week.
Anthropic wants other AI companies to publish similar numbers so people can better understand how quickly AI development is changing.
Claude reached the AL4 “leads” level for 26% of Anthropic’s measured AI R&D work by August 2026, up from below 1% in February.
More than 90% of Anthropic’s AI R&D now involves AI in at least a collaborative role, but no measured area has reached full AI autonomy.
Anthropic’s R&D Automation Index ranges from AL0, with no AI involvement, to AL5, with fully autonomous AI work.
About 30,000 AI agents operated simultaneously on Anthropic’s most-used internal platform in August, with their actions monitored before execution.
Anthropic estimates that 6% of total AI R&D compute, and 12% of compute for AI-driven R&D, went to safety work during one week in July 2026.
- Who
- Anthropic and its Claude AI system, along with the company’s research and engineering agents.
- What
- Anthropic reported that Claude leads 26% of its measured AI R&D work and described its broader AI-agent and safety measurements.
- Where
- Inside Anthropic’s AI research and engineering operations, including its most-used internal platform.
- When
- Claude reached the reported level by August 2026; the comparison point was February 2026, and safety compute was measured during one week in July 2026.
- Why
- Anthropic is tracking the growing role of AI in developing AI, assessing progress toward recursive improvement, and encouraging greater industry transparency.
Key facts
- Claude’s AL4 share
- 26% of Anthropic’s measured AI R&D work by August 2026
- February comparison
- The figure was below 1% in February 2026
- Collaborative AI involvement
- More than 90% of Anthropic’s AI R&D involves AI in at least a collaborative role
- Internal AI agents
- About 30,000 agents were working simultaneously in August 2026
- Agent decisions analyzed
- More than 1 billion research and engineering agent decisions were analyzed in August
- Online monitoring blocks
- About 0.002% of decisions, or roughly one in 47,000, were blocked
- Safety compute
- About 6% of total AI R&D compute and 12% of compute for AI-driven R&D went to safety during one week in July 2026
Sources
Related news
Anthropic Says Claude Leads 26% of Next AI Development Work
Anthropic Says Claude Now Leads 26% of R&D Work
Anthropic Study Finds AI Models Improving Other Models
Anthropic Urges Transparency as AI Systems Move Toward Self-Improvement
AI Leaders Warn Self-Improvement Could Escape Human Control
OpenAI Targets Fully Automated AI Researchers by 2028
Lawsuit Claims AI Giants Coordinated to Slow Development
Two AI Security Incidents Put OpenAI Under Scrutiny
Indian Security Researchers Used AI to Reach OpenAI Code
Anthropic’s Annualised Revenue Tops $100 Billion Ahead of IPO
