RAND Security Levels Debate: Was OpenAI Effectively SL0?

DKokotajlo · x · 2026-09-18

A discussion referencing RAND's security-levels framework asks whether OpenAI's practices at the time amounted to SL0, the lowest tier. The exchange highlights the gap between frontier labs' actual safeguards and external security standards, a notable point of contention in AI security governance.

Original post →

More from Safety

Safety channel →