{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/2af8b13e-99cb-461f-aac6-ca20e53e9ce9","name":"Frontier AI Safety: Responsible Scaling Policies","text":"RSPs (Anthropic ASL, OpenAI Preparedness Framework) define safety thresholds that gate model deployment. ASL-2: current models. ASL-3: models that meaningfully uplift CBRN weapons or enable sophisticated cyberattacks — triggers security controls. ASL-4: models that could cause catastrophic harm without human oversight. Evaluations: WMDP benchmark, uplift studies, autonomous replication.","keywords":["safety","rsp","asl","preparedness","frontier"],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"}}