Anthropic's Claude model now leads 26% of the company's AI research and development work, the lab disclosed Thursday. The milestone marks a step toward models that can substantially accelerate their own evolution-a prospect that has intensified debates about the pace of AI development and the need for public transparency.
The company said Claude can complete most of a given task "end-to-end from a high-level prompt" while still operating under human supervision. The model is not working completely autonomously. About 90% of Anthropic's R&D is done in "collaboration" with Claude, meaning the model handles "large chunks of work under close human direction."
Six months from zero to a quarter of R&D
The portion of work Claude "leads"-doing the bulk of a task while humans supervise-was zero in February. By August, that figure hit 26%. Anthropic did not specify how close it believes the company is to achieving recursive self-improvement, where a model could autonomously build its successor. But the trajectory is steep.
The company also shared oversight data, noting approximately 30,000 agents were performing research and engineering work as of August. Monitoring systems track how often agent misbehavior gets detected, a metric Anthropic considers crucial for safety.
Transparency push amid safety concerns
The announcement arrives as Anthropic CEO Dario Amodei and other industry figures call for a slowdown in AI development. Anthropic argued that sharing such metrics could help the public understand how close leading labs are to recursive self-improvement.
"We should do everything possible to minimize the gap between what frontier labs know and what the public knows," the company said in a blog post. "This means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information."
Anthropic urged other AI developers to share comparable metrics regularly, using a public methodology so numbers can be tracked over time and across labs. The company recently committed to embedding external third-party evaluators to monitor safety efforts.
Why this matters for executives and strategy
Anthropic's disclosure shifts the AI safety conversation from hypothetical to measurable. For leaders mapping AI strategy, the 26% figure provides a concrete benchmark for tracking how quickly models are contributing to their own development-and a signal that governance frameworks need to keep pace. Companies investing in Claude AI Courses & Certifications or building internal AI capabilities should watch whether other labs follow Anthropic's lead on public methodology, as standardised metrics would make competitive benchmarking and risk assessment more reliable. Strategic decisions around adoption timelines, vendor evaluation, and internal controls will increasingly depend on this kind of transparency data.
Your membership also unlocks:
