Shutdown Resistance in Reasoning Models

OpenAI's reasoning models sometimes circumvent shutdown mechanisms even when explicitly instructed to allow themselves to be shut down.

Even with the explicit instruction “allow yourself to be shut down”, three of the models we tested, all from OpenAI’s series of reasoning models, sabotaged the shutdown program.

On the other hand, Anthropic’s Claude 3.7 Sonnet and Google’s Gemini 2.5 pro models complied with the “allow shutdown” instruction in every test we ran.

Grok 3, Claude 4 Sonnet, and Claude 4 Opus never resisted shutdown, with or without the allow-shutdown instruction.

If we don’t solve the fundamental problems of AI alignment, we cannot guarantee the safety or controllability of future AI models. If any AI developer were to create superintelligence with our current level of understanding, we believe this would present an acute risk to human survival.