Cohere CEO Criticizes Anthropic's Proposal for AI Safety Cartel

Drawing parallels to historical regulatory failures like the US credit rating agencies and the EU automotive sector, Gomez contends that safety-focused regulations often inadvertently protect incumbents and stifle competition. He emphasizes that while AI requires guardrails, the current proposal lacks transparency, democratic consultation, and scientific evidence-based criteria. Instead of a closed-door agreement among competitors, he advocates for an open, international framework where risk definitions are established by diverse experts and governments, not by the companies being regulated.

Gomez proposes four pillars for a better regulatory approach: an evidence-based risk framework developed internationally, mandatory transparency from developers, testing scoped by specific capabilities rather than company size, and independent assurance mechanisms similar to those in aviation or finance. He insists that safety standards must be applied equally to all entities based on their capabilities, ensuring that the most consequential technology in human history is not controlled by a small group of commercially aligned executives.