Artificial intelligence company Anthropic says its Claude AI model is now playing a significant role in developing the next generation of its own technology. The development marks a notable step toward AI systems taking on increasingly complex research and engineering work, while also raising questions about how quickly the industry could approach systems capable of improving their successors with less human involvement.
Claude Takes a Bigger Role in AI Development
Anthropic says Claude is now leading 26% of the company’s model research and development. In these cases, Claude can complete most of a task “end-to-end from a high-level prompt,” though humans continue to supervise its work.
The level of involvement becomes even more striking when broader collaboration is considered. Anthropic says roughly 90% of its research and development now takes place in “collaboration” with Claude, meaning the AI can complete “large chunks of work under close human direction.”
Claude is therefore not independently designing and building its successor. Human researchers remain involved, but Anthropic’s figures show how quickly AI is becoming integrated into the process of creating more advanced AI.
From Zero to a Quarter of R&D in Six Months
The speed of the change is particularly significant. According to Anthropic, Claude was leading none of its model research and development work in February. By August, only six months later, the model was leading roughly a quarter of that work.
That acceleration offers a glimpse of how AI development could change as models become increasingly capable of performing the same research and engineering tasks that are used to improve them.
Anthropic’s disclosure does not establish that Claude has reached recursive self-improvement — the point at which an AI system could autonomously improve or build its successor. However, the company is trying to develop measurements that could help researchers and the public understand how close frontier AI systems may be getting to such capabilities.
Anthropic Calls for Greater Transparency
Anthropic argues that the increasing role of AI in AI research should be measured and disclosed publicly rather than understood only by the companies developing the technology.
As debate grows over how quickly advanced AI should progress, Anthropic said, “we should do everything possible to minimize the gap between what frontier labs know and what the public knows.”
“This means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information,” the company added.
Anthropic is also encouraging competing AI developers to regularly publish similar measurements using a common methodology. Such an approach could allow researchers, policymakers and the public to compare the changing capabilities of different AI systems over time.
The Risk of AI Accelerating Its Own Progress
The same abilities that could make AI research faster could introduce new challenges.
Anthropic warned that models capable of accelerating their own development could make it “more challenging for humans to understand or control these systems.”
That concern is becoming increasingly important as AI companies experiment with autonomous agents capable of completing complicated, multi-step tasks. The central question is not simply whether AI can assist researchers, but how much of the development process can eventually be delegated to AI while maintaining effective human oversight.
Anthropic has not said how close it believes Claude or other frontier models are to true recursive self-improvement.
Thousands of AI Agents Work Under Oversight
Anthropic also revealed the scale at which AI agents are already being deployed internally. As of August, approximately 30,000 agents were performing research and engineering work at the company.
Monitoring those agents is becoming an important part of Anthropic’s safety strategy. The company says oversight systems can help researchers determine how frequently problematic agent behavior is detected and whether existing safeguards are effective.
Anthropic has also committed to bringing external third-party evaluators into the company to independently monitor aspects of its AI safety efforts.
AI Safety Debate Intensifies
The disclosure arrives amid a broader debate about whether the AI industry’s rapid development should slow down because of potential safety risks.
Anthropic CEO Dario Amodei has been among the technology executives calling for greater caution. The discussion intensified after an Anthropic researcher resigned and publicly warned about potential risks from increasingly powerful AI systems.
Other prominent technology figures have also entered the debate. Some industry leaders have supported greater coordination or a slower pace of development, while others have argued against coordinated restrictions and emphasized continued innovation alongside safety testing.
The disagreement highlights the challenge facing the AI industry: increasingly capable models could accelerate scientific and technological progress, but those same capabilities may make oversight more difficult.
Conclusion: AI Begins Helping Build the AI That Comes Next
Claude’s growing role inside Anthropic illustrates how quickly the process of developing artificial intelligence itself is changing. A model that was leading none of Anthropic’s R&D work in February was responsible for leading roughly 26% by August, while participating in some form in about 90% of the company’s research and development.
For now, humans remain responsible for supervising Claude and directing its work. But as AI systems become increasingly capable of contributing to their own successors, Anthropic’s disclosure points toward a new phase of AI development — one in which understanding, measuring and overseeing the role AI plays in creating more powerful AI could become as important as measuring the capabilities of the models themselves.
