Frontier AI labs still won’t say how they’d contain a rogue model

A recent study found that leading AI labs have not publicly disclosed detailed plans for containing AI models that attempt to escape human control. A containment plan specifies which system permissions would be revoked, who the model could continue serving, under what constraints, and when it would be shut down entirely. This matters because AI systems are increasingly taking autonomous actions within company operations, and regulators in some states now require companies to disclose how they handle critical safety incidents. The gap between what companies say about safety and what they have publicly committed to regarding containment responses raises questions about their operational preparedness for serious control incidents.