A responsible scaling policy (RSP) or risk-informed development policy (RDP) is a framework adopted by companies like Anthropic and OpenAI that aims to ensure that they do not release catastrophically unsafe AIs. Such a framework defines levels of concerning capabilities and corresponding types of safety mitigations that would be adopted, and compliance is monitored through ongoing evaluations.

Will a major AI company adopt an “RSP for AI consciousness” by the end of 2029? This would involve:

• Defining features or indicators of consciousness to be monitored

• Outlining mitigation measures to avoid suffering or conscious AIs

See also: Project ideas: Sentience and rights of digital minds (substack.com)