Anthropic founder Dario Amodei introduced the Responsible Scaling Policy (RSP) at the UK Artificial Intelligence Safety Summit. This policy aims to ensure the safety and controllability of AI systems, preventing potential risks and misuse. RSP consists of two main components: the ASL system and scaling curves. The ASL system is modeled after the Biosafety Level (BSL) system, with each level having corresponding safeguards. ASL-1 represents models with almost no risk, such as professional AI for chess; ASL-2 represents the current stage, where models have broad risks but do not yet exhibit truly dangerous capabilities; ASL-3 is when AI models become operationally dangerous in the CBRN (chemical, biological, radiological, and nuclear) fields; ASL-4 represents an escalation of catastrophic misuse risk, triggered when AI systems approach human-level autonomy or become a major source of at least one serious global security threat (such as biological weapons). Scaling curves are used to test the intervals of these dangerous capabilities to ensure that models with dangerous capabilities are not created blindly. As the scale and capabilities of AI systems increase, safeguards need to be upgraded accordingly.