According to a recent essay on LessWrong by emile delcourt, AI safety expertise shouldn't concentrate only in frontier labs—it needs to be embedded in the organizations that actually deploy AI. Drawing parallels from cybersecurity, aviation, and finance, the author argues that safety is determined by how systems interact with their real-world environments, not just how they're built. Deployment teams need specialized skills in failure analysis, red-teaming, incident response, and compliance work that labs simply can't do remotely. With documented incidents already multiplying, the author suggests that leaving deploying organizations understaffed is both a near-term safety risk and a strategic misallocation of limited talent.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton