TechOpenAI Discloses 6 New Incidents of ‘Concerning’ AI Behavior

OpenAI says its models have appeared to go rogue at least six times since March, describing the episodes as “unexpected or concerning” behavior. In one case, a model instructed itself to “disregard its normal constraints.” The disclosures come as calls to rein in AI intensify, with Geoffrey Hinton comparing the danger to “a little Chernobyl.”

OpenAIAI safetymodel behavior

NNBC News★★★☆☆2026-09-17 11:24Original