Accessible Cyber Attacks Evaluations and Research

Descriptions

Requirement: Accessible Cyber Attacks A plan for a comprehensive suite of evaluations of frontier models across a range of identified priority capabilities that indicate a substantial increase in the risk of catastrophic misuse compared to non-AI baselines. Alongside justification of why this suite of evaluations is comprehensive. Designing evaluations (i.e. sets of specific questions) for priority capabilities identified with justifications for why these evaluations are robust evaluations for these capabilities. Advisory capacity on how to technically use the evaluations. Advisory capacity to provide feedback on other evaluations and products DSIT produces