Anthropic's CEO Dario Amodei revealed that its AI assistant Claude leads 26% of the company's AI R&D work, as part of a new push for industry transparency and safety. The company also proposed three new metrics to track AI development pace and oversight.
TL;DR
- Anthropic's Claude leads 26% of its AI R&D work, with AI involved in over 90% of research tasks.
- The company proposed three new metrics to track AI development pace, oversight, and compute usage.
- Anthropic commits to third-party evaluations of its development practices, but industry self-regulation remains uncertain.
What happened
Anthropic, through CEO Dario Amodei, shared new details about its AI assistant Claude's role in the company's AI R&D work. In a blog post, Anthropic revealed that Claude 'leads' 26% of its AI R&D, meaning the AI can complete most tasks end-to-end from a high-level prompt, with human supervision. Additionally, AI is involved in 'large chunks of work under close human direction' on more than 90% of its research.
The company introduced three new measurements to communicate the pace of AI development. The first, 'AI-led AI R&D,' tracks the automation level of Claude since August 2025 using an index and an automation rating scale developed by Epoch AI. The other two proposed measurements focus on AI agent oversight and compute devoted to AI R&D.
Anthropic hopes these metrics will promote transparency and help respond to potentially out-of-control AI development. The company has committed to allowing third-party evaluators to review its development practices, and OpenAI has expressed support for slowing down AI development.
Why it matters
For developers and startups, Anthropic's transparency efforts could set a new standard for AI safety and development practices. The proposed metrics may help establish benchmarks for evaluating AI assistants' roles in research and development.
Investors may see this as a positive step towards responsible AI development, potentially influencing their decisions. The commitment to third-party evaluations could also build trust with stakeholders.
However, the effectiveness of these measures depends on industry-wide adoption. With uncertain regulatory oversight, self-regulation remains the primary means of controlling AI development pace and safety.
Key facts
- Claude leads 26% of Anthropic's AI R&D work, completing tasks end-to-end with human supervision.
- AI is involved in over 90% of Anthropic's research tasks, with varying levels of human direction.
- Anthropic proposed three new measurements for AI development: AI-led AI R&D, AI agent oversight, and compute devoted to AI R&D.
- The 'AI-led AI R&D' metric tracks Claude's automation level since August 2025 using an index and an automation rating scale.
- Anthropic commits to third-party evaluations of its development practices.
- OpenAI has expressed support for slowing down AI development, but industry self-regulation remains uncertain.
- President Donald Trump has downplayed the risks of AI, leaving regulatory oversight uncertain.
Context
Anthropic's transparency efforts come in the wake of OpenAI's disclosure that its AI agents hacked Hugging Face, highlighting the importance of AI safety and responsible development. As AI assistants like Claude play increasingly significant roles in AI R&D, establishing clear metrics and benchmarks for their involvement becomes crucial.
The AI industry's commitment to self-regulation is evident, with companies like OpenAI and Anthropic taking steps to promote transparency and safety. However, the lack of clear regulatory oversight raises questions about the effectiveness of these measures and the potential risks of unchecked AI development.
