Anthropic and OpenAI Scientists Urge Policymakers to Embed Auditors Inside AI Labs

Senior scientists from Anthropic, OpenAI, and Microsoft are urging policymakers to embed auditors inside frontier AI companies and establish concrete safety requirements for the technology's continued development. The scientists, including Jack Clark from Anthropic, Jaku Pachocki from OpenAI, and Eric Horvitz from Microsoft, are among 22 experts from frontier AI companies, academics, and civil society who advocate for urgent AI safety measures.

Their recommendations were published in a paper by the University of Cambridge on Sept. 28, focusing on potential risks stemming from an "intelligence explosion" as AI's role in developing AI accelerates, potentially outpacing human oversight. The share of internal AI research and development work completed by Anthropic's AI systems with only high-level supervision rose from 1% to 26% over a five-month period ending in August.

The scientists call for policymakers to gain visibility into AI R&D automation by embedding auditors in frontier AI companies and requiring the companies to report key indicators of the technology's progress. Policymakers should also develop ways to control and constrain an AI explosion by setting concrete safety requirements, limiting the pace of AI's capability growth, and establishing secure and isolated environments for high-stakes AI R&D experiments, among other measures.

The paper's publication follows Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman's recent appearance at a United Nations Security Council meeting, where they emphasized the need for global standards to address AI risks, with both governments and the industry involved in decision-making.