Intolerable Risk Threshold

A commitment made by AI companies at the 2024 Seoul AI Safety Summit to define severity levels of risk, such as large-scale harm to public safety, that their frontier models must never be allowed to cross, along with the mitigations needed to stay under that threshold.