The company says Claude now leads 26% of its internal AI research work, up from under 1% in February
Anthropic believes Claude will surpass human researchers within nine months. The company is no longer talking about AI as a research assistant. It is talking about AI as the one running the lab.
The supporting data is striking. On September 17, 2026, Anthropic disclosed that its Claude models now lead 26% of its internal AI research and development work. In February 2026, that figure was under 1%.
Anthropic grades AI involvement in research on a scale of automation levels. The level that matters here is AL4, which the company describes as the stage where the AI leads the work.
By that measure, Claude led 26% of Anthropic’s R&D efforts as of the September disclosure. More than 90% of the R&D tasks Anthropic measured involved AI collaboration at AL3 or higher as of August 2026.
Anthropic reported no fully autonomous work, which it classifies as AL5. Approximately 30,000 AI agents were active on Anthropic’s internal platform as of August 2026.
Anthropic projects that Claude could lead around 80% of its internal AI R&D at the AL4 level by the end of 2026, provided the current trend continues.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
In April 2026, agents running Claude Opus 4.6 closed 97% of the performance gap on an alignment problem compared with human researchers. They did it in seven days.
Experiments in early 2026 showed Claude agents outperforming human researchers on alignment and safety tasks, but those gains were sometimes not replicable in production models.
In tests conducted in August 2026, Claude designed protein binders and posted higher success rates than human experts. Protein binders are molecules engineered to latch onto specific biological targets, a core building block in drug discovery and biotech research.
If Claude leads around 80% of internal R&D by the end of 2026, as the company projects is possible, the role of a human researcher there shifts toward review, direction and judgment calls.
Anthropic’s own data shows that some alignment gains did not carry over to production models. If AI increasingly leads the research, catching those gaps depends on humans who are doing less of the hands-on work. The company’s emphasis on human oversight reads as an acknowledgment of exactly that tension.
Investors and stakeholders tracking the AI sector have a few concrete markers to watch. The first is whether Anthropic’s next disclosure shows the AL4 share moving toward the projected 80%. The second is whether any work crosses into AL5, the fully autonomous tier that has not yet appeared in the company’s measurements. The third is whether AI-led safety improvements start holding up consistently in production models rather than only in controlled experiments.