Technology

Anthropic says Claude AI model leads quarter of its research and development

The company disclosed that its system is helping build its successor while remaining under human supervision.

Updated Sep 18, 2026 · 19:14 UTC6 reports

The short version

Anthropic announced that its AI model Claude now leads 26% of internal research and development work, up from none in February.[ABC News]
About 90% of the lab's research involves Claude in some collaborative capacity, though humans continue to supervise tasks.[ABC News]
Roughly 30,000 agents were conducting engineering and research tasks at the firm as of August.[ABC News]
Anthropic noted it remains unclear how close developers are to recursive self-improvement, warning that rapid self-advancement could impair human oversight.[ABC News]

Claude leads 26% of Anthropic's model research and development tasks, allowing it to complete tasks from high-level prompts under human supervision.[ABC News]
Claude is not fully autonomous, though approximately 90% of Anthropic's research and development involves close collaboration with the system.[ABC News]
Anthropic reported having roughly 30,000 agents performing research and engineering tasks by August.[ABC News]
An Anthropic researcher resigned with a warning regarding the societal threats posed by rapid artificial intelligence development.[ABC News]
Anthropic agreed to embed independent third-party evaluators to oversee its safety initiatives.[ABC News]

Anthropic's disclosure did not establish how close the organization is to achieving recursive self-improvement.[ABC News]
Sources present differing baselines for Claude's lead work share prior to the surge, citing none in February versus 1% in March.[ABC News · Business Insider]

Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.