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

A new study says leading AI labs have disclosed few concrete plans for containing rogue models if advanced systems behave unexpectedly or dangerously. The findings raise concerns about whether companies are prepared for emerging risks as AI capabilities continue to accelerate.

The lack of public containment strategies points to a growing gap between AI deployment and safety governance. As frontier models become more autonomous and unpredictable, pressure may build for standardized disclosure, third-party audits, and regulatory requirements around incident response.