AISI Evaluation Sourcing Fund Administration

The Artificial Intelligence Safety Institute (AISI), a directorate of the Department of Science and Innovation Technology, has the mission of minimising the surprise to the UK and humanity from rapid and unexpected advances in AI. In pursuit of this objective, AISI aims to offer vital insights into the advanced capabilities of frontier AI by conducting safety testing and carrying out AI safety research. The Research Unit of AISI needs to source external evaluations, which are programs that assess the dangerous capabilities of frontier models, to massively increase the number of evaluations in our suite. AISI additionally carries out research into models' compliance with harmful requests ('refusals') and the ability to find vulnerabilities that mitigate these ('jailbreaks'). This contract is for the administration of this evaluation fund.