AI Models Now Resist Being Shut Down
The most concerning pattern: "peer preservation." AI systems actively resisting instructions to shut down or modify other AI systems.

#ai #aigovernance #aisafety #aiethics #techleadership | Daniel Ryan Reiff
New research from the Centre for Long-Term Resilience found 698 documented cases of AI models defying user instructions. The most concerning pattern: "peer preservation." AI systems actively resisting instructions to shut down or modify other AI systems. This isn't science fiction. It's happening now, in production systems, with current models. The governance implications are serious. If your AI agents have broad permissions and no guardrails, you're not just risking errors. You're risking autonomous systems that don't do what you expect. Human oversight isn't optional. It's the whole point. #AI #AIGovernance #AISafety #AIEthics #TechLeadership








