Anthropic, OpenAI, and Google have been meeting since July to design an industry-led AI standards body, The Information reported on 13 September. The talks sit beside Amodei’s public plan for common safety standards and embedded evaluators.
An industry body would let the largest labs write shared rules on evaluation, incident reporting, and release pacing without waiting for a finished statute. Supporters call that speed. Critics call it regulatory capture: the firms with the most to lose from hard limits designing the limits themselves.
Microsoft added a parallel track over the weekend. Satya Nadella published a Code of Conduct for Microsoft’s MAI models and said the company welcomes deliberate pacing and broader representation across countries and academic fields—not only a closed circle of labs.
Two safety researchers also left Anthropic and Google DeepMind for METR, the independent evaluation group, on 12 September. That personnel flow is a practical test of whether “embedded evaluators” will be independent in fact or only in branding.
Key takeaway The three most prominent U.S. model developers have been quietly drafting an industry standards vehicle while their CEOs argue in public about pacing.
Photo: Unsplash / conference table. Sources: The Information via AI Weekly, The Register, Techmeme/Nadella coverage.
