Frontier Model Forum introduces preliminary jailbreak risk-rating framework for cyber threats

The Frontier Model Forum introduces a preliminary framework for rating successful jailbreaks of AI models by their potential risk.

The framework draws on expertise across FMF members and is described as a shared method for assessing jailbreaks as AI cyber capabilities evolve.

The forum's announcement, posted on X, links to its own page describing the framework. Only the forum's description of the framework is available here, so details of the rating criteria and how the method is applied are not reported.