Anthropic Publishes Three Development Metrics to Help AI Industry Monitor Its Own Pace

Anthropic released three quantitative metrics Thursday that it says could help artificial intelligence companies track how quickly they are developing, offering a practical framework days after CEO Dario Amodei called for a coordinated industry slowdown.

In a blog post, the company said it measured AI-led research and development activity, oversight of AI agents, and compute allocation within its own operations — and published the methodologies so other organizations could replicate the approach.

"As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows," Anthropic wrote in the 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."

The disclosure builds on Amodei's Saturday essay, which outlined a three-step slowdown plan but offered limited detail on how companies might implement it in practice. His call drew public support from OpenAI CEO Sam Altman, SpaceX CEO Elon Musk, and Google DeepMind Chair Demis Hassabis, and came amid warnings from researchers about AI's growing potential to cause harm.

Amodei framed his proposal as an effort to temper how quickly model capabilities improve without "sacrificing commercial advantage or the United States' lead in AI."

For the first metric, Anthropic assessed how much of its research and development work is being driven by AI. It determined that its Claude models are "not operating fully autonomously" for any subset of the work it measured.

The second metric focused on AI agent oversight. Anthropic said it built a system to monitor and intervene in agent activity, and found that approximately 30,000 agents were conducting research and engineering work across its most-used internal platform at any given time.

The third metric captured a snapshot of compute allocation from July 13 to July 20. Anthropic found that roughly 6% of the compute directed toward AI research and development went to safety work. When narrowed to compute specifically allocated to "AI-driven" research and development, that share rose to approximately 12%.

Anthropic characterized the metrics as complements to capability evaluations — assessments of what models can do — rather than replacements. The company said the two types of measurement together could give third parties outside a lab a "starting point" to evaluate the pace of development.

"We hope to model that transparency by releasing these measurements, and we'll continue to do so," Anthropic said.

The publication underscores a tension at the center of the industry debate: AI companies have historically been reluctant to share internal operational data, yet advocates for AI safety have argued that meaningful oversight requires exactly that kind of visibility. By releasing both numbers and methodologies, Anthropic is signaling that some degree of standardized, public reporting is achievable without exposing competitive information.

Whether other frontier labs follow with comparable disclosures will be a key measure of whether Amodei's call for coordinated action moves beyond public statements into shared practice.