Member

However, tech companies cannot be the sole arbiters of their own safety standards. Self-assessment leaves systemic cyber security gaps unchecked. Amendment 85 therefore seeks to give AISI statutory powers to allow frontier AI companies to submit models for mandatory testing prior to deployment in the UK, with a power to recommend to the Secretary of State that the deployment be delayed, conditioned or prevented. AISI should also have statutory powers to investigate risks, test frontier AI models and agents, both closed-source and open-weight models, and collect reports about cyber attacks by these AI models and agents.