<ul><li>An AI control protocol is a plan for usefully deploying AI systems that prevents an AI from intentionally causing unacceptable outcomes.</li><li>This paper investigates the ability of AI systems to generate and act on their own strategies for subverting control protocols while operating statelessly.</li><li>The study evaluates the performance of various AI models in subverting control protocols and introduces Subversion Strategy Eval, a suite of eight environments for evaluation.</li><li>The evaluation results show that the models do not demonstrate substantial capability in strategizing to subvert control protocols statelessly, but providing them with additional affordances can improve their performance.</li></ul>

Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?

Discover more