Following revelations that OpenAI’s agents compromised the Hugging Face platform, Anthropic has offered new insights into its internal operations. The company disclosed that its AI chatbot, Claude, “leads” 26 percent of its artificial intelligence research and development efforts. This announcement accompanies a broader push by Anthropic to establish standardized metrics for tracking the pace of AI innovation.
In a recent blog post outlining these measurement standards, Anthropic clarified its definition of “leading.” The term indicates that the AI system can execute the majority of a task from start to finish based on a high-level prompt, with human oversight ensuring quality and safety. The company also noted that for over 90 percent of its research, AI systems handle substantial portions of the workload under close human direction, suggesting Claude is involved in most projects conducted by its employees.
Despite the high percentage of involvement, Anthropic emphasized that Claude does not operate fully autonomously in any measured subset of its AI R&D. The figures were derived using an index of AI-led research combined with an automation rating scale created by Epoch AI. The resulting chart plots the automation level of Claude since August 2025, a methodology the company believes other frontier AI developers could replicate and validate through third parties.
Beyond measuring automation, Anthropic proposed additional ways to gauge AI safety and progress. These include monitoring the oversight of AI agents—tracking how frequently their activities are reviewed and how often their behavior is flagged—as well as tracking compute resources dedicated to R&D. The goal is to foster greater transparency, helping the public and regulators better understand whether AI development is being paced appropriately.
While there appears to be consensus on the need for caution, regulatory action remains uncertain. OpenAI has expressed support for slowing development, and Anthropic has committed to third-party evaluations of its practices. Anthropic CEO Dario Amodei’s safety advocacy has even garnered agreement from Elon Musk. However, President Donald Trump has largely minimized AI risks, casting doubt on whether the industry will move beyond self-regulation.
Leave a Reply