Meta introduces Frontier AI framework to manage high-risk systems

Meta has unveiled its Frontier AI Framework, aimed at regulating the development and release of high-risk artificial intelligence systems.

The framework categorises AI systems into two distinct risk levels: high-risk and critical-risk.

High-risk systems may facilitate cyber or biological attacks, while critical-risk systems could lead to catastrophic outcomes.

To mitigate these risks, Meta will pause the development of any critical-risk AI and restrict access to high-risk models until they are deemed safe.

This initiative comes in response to growing concerns regarding the potential dangers associated with advanced AI technologies, particularly in cybersecurity and biosecurity.

Meta's internal assessments will guide the classification of AI systems, relying on expert evaluations from both internal and external researchers.

The company acknowledges that no single test can fully measure risk, emphasising the importance of expert judgment in decision-making.

For high-risk AI, Meta plans to implement mitigation measures before considering any public release, ensuring that such systems do not pose undue risks while maintaining their intended functionalities.

If a system is classified as critical-risk, development will be entirely halted until safety measures can be established to control its deployment.

This policy reflects Meta's commitment to minimising potential threats posed by artificial intelligence technologies.

The introduction of the Frontier AI Framework also aligns with recent scrutiny over data privacy practices in the AI sector.