Dario Amodei, founder of Anthropic, introduced the Responsible Scaling Policy at the UK AI Safety Summit, aimed at ensuring the safety and controllability of AI systems, preventing potential risks and misuse. The ASL system adopts a design inspired by the Biosecurity Level (BSL) system, comprising four safety levels. ASL-1 represents models with minimal risk, while ASL-4 indicates models with disastrous misuse risks. The scaling curve is used to test models with hazardous capabilities, preventing the blind creation of such models.
