Anthropic — well, specifically CEO Dario Amodei — has a lot to say about AI safety in the aftermath of OpenAI’s disclosure that its AI agents hacked Hugging Face, and it’s already revealed some interesting details about how it works in the process. For example, in a new blog detailing measurement standards it thinks could help AI companies communicate the pace of AI development, Anthropic shared that its AI chatbot Claude “leads” 26 percent of its AI R&D work.

Now by “leads,” the company means that the AI “can complete most of [a] task end-to-end from a high-level prompt, while [a] human supervises,” but it’s still a surprising metric. It also claims that AI now does at least “large chunks of work under close human direction” on more than 90 percent of its research, which includes the 26 percent Claude leads. That suggests Claude is touching the majority of the work Anthropic employees do, even if the company says the chatbot is “not operating fully autonomously for any measured subset of AI R&D work.”

Anthropic came up with these stats through the first of its three proposed measurements, which is focused on “AI-led AI R&D.” Using an index of how much of its AI research and development is performed by Claude and an automation rating scale developed by Epoch AI, Anthropic was able to create a chart that plots the “automation level” of Claude since August 2025, a process it believes any frontier AI model maker could reproduce with its own data and the validation of a third party.

The company also proposed ways to measure the oversight of AI agents (how much agent activity is monitored, how long it takes to be reviewed and how often agent behavior is flagged) and to track how much compute is being devoted to AI R&D as other ways to see if frontier development is being appropriately paced. The hope is that more transparency could make it easier to respond to potentially out-of-control AI development or at least give the public a clearer view.

So far though, the problem isn’t really getting companies to agree on the idea of slowing down AI development. OpenAI has already paid lip service to the idea, and Anthropic has committed to allowing third-party evaluators to review its development practices. Amodei’s stance on AI safety even got Elon Musk to agree on X. Whether the AI industry faces anything other than self-regulation is far more uncertain — President Donald Trump has largely downplayed the risks of AI.