Inside the Explosive Growth of AI Safety Research: METR, Redwood, OpenAI, and Anthropic

A deep-dive piece from The Verge surveys the rapidly expanding AI safety research landscape, profiling organizations including METR, Redwood Research, OpenAI, and Anthropic and the divergent technical approaches they are pursuing. The piece documents how safety has shifted from a fringe academic concern to a well-funded, institutionally competitive field with distinct methodological camps — from interpretability and mechanistic analysis to red-teaming and scalable oversight. For developers building on top of frontier models, understanding the safety research ecosystem matters because it directly shapes what capabilities get gated, how model behavior is constrained, and what disclosure obligations may emerge. The coverage also contextualizes the OpenAI misalignment framework release and Anthropic's internal Claude usage disclosures as part of a broader institutional push toward formal safety accountability. Teams integrating agentic AI into products should track these developments as early indicators of where the regulatory and technical floor will land.