Meta Introduces Framework to Limit Release of High-Risk AI Systems

The company’s new policy outlines measures to address potential catastrophic risks posed by advanced AI models.

Meta's Frontier AI Framework categorizes AI systems into 'high risk' and 'critical risk' based on their potential to cause harm.
High-risk AI systems could facilitate cyberattacks or aid in the development of chemical and biological weapons, but with mitigable risks.
Critical-risk systems are defined as those that could lead to catastrophic outcomes, such as fully automated cyberattacks or the proliferation of biological weapons, with no viable mitigation strategies.
Meta states it will halt development and prevent the release of critical-risk AI models until they can be deemed safe for deployment.
The policy highlights Meta’s response to criticism of its open AI development strategy and contrasts with competitors like China’s DeepSeek, which lacks similar safeguards.