Anthropic says Claude is helping build the next generation of AI models

Friday 18 September 2026 - 23:42

By: Azzat Manal

Anthropic says Claude is helping build the next generation of AI models

Anthropic has disclosed new figures showing how rapidly artificial intelligence is becoming involved in the development of more advanced AI systems, with its Claude model now leading a significant share of the company’s own research and development work.

According to measurements released by Anthropic, Claude was responsible for leading 26% of the company’s AI research and development activities as of August 2026. The figure represents a sharp increase from less than 1% earlier in the year and reflects the growing role of AI systems in tasks traditionally carried out by human researchers and engineers.

Anthropic said Claude is not yet operating fully autonomously in any of the areas measured. Instead, the model can complete substantial portions of certain tasks under close human supervision. More than 90% of the company’s AI research and development work involved some form of collaboration between humans and AI during the period assessed.

The company also reported that around 30,000 AI agents were carrying out research and engineering work on its most widely used internal platform in August. Anthropic said the agents’ activities are subject to monitoring systems designed to detect or block potentially dangerous actions, with every action passing through an online monitoring system before execution.

Anthropic has presented these figures as part of an effort to provide greater transparency about the pace at which AI is being incorporated into the development of future models. The company argues that tracking how much AI contributes to AI research could help researchers and the public understand how close the industry may be to what is known as recursive self-improvement, a hypothetical stage in which an AI system could independently design and develop increasingly capable successors.

Anthropic has stressed that such a scenario has not yet been reached and that recursive self-improvement is not inevitable. However, the company has acknowledged that increasing automation could make advanced AI systems more difficult for humans to understand or control, particularly if the systems become capable of handling larger parts of their own development process.

The disclosure comes amid a broader debate over the pace of AI development and the safeguards needed as models become more capable. Anthropic CEO Dario Amodei recently called on AI companies to slow the development of increasingly powerful systems to allow additional safety measures and oversight to keep pace with technological progress.

Anthropic has also been expanding research into AI alignment, including experiments in which Claude-powered systems are used to study and mitigate potential alignment failures. The company says such work is intended to ensure that safety research develops alongside the capabilities of increasingly powerful AI models.

The growing role of AI in AI development is therefore creating a new dynamic for the technology industry. Human researchers continue to define objectives, supervise systems and assess their results, while AI agents increasingly contribute to coding, experimentation, evaluation and other stages of model development.

Anthropic’s latest measurements provide a snapshot of that transition rather than evidence that AI systems are independently developing themselves. The company says regular reporting and independent evaluation could help governments, researchers and the public better understand how quickly this relationship between humans and AI is changing.