The UK AI Safety Institute (AISI)'s pre-deployment evaluation model is under pressure; the author suggests expanding AI assurance functions
Britain’s AI Assurance Approach Needs a Second Act
AI summary
The UK AI Safety Institute (AISI)'s pre-deployment evaluation model, which relies on voluntary cooperation from frontier labs, is under pressure, and the author argues it needs to enter a new phase. The institute has no authority to compel companies to share models; Politico reported that the White House once asked labs to pause providing new models to it, and the pause was lifted after 17 days; over three years its pre-deployment assessments have found dozens of vulnerabilities, and 10 countries have established similar institutions. The author suggests it deepen the UK AI assurance ecosystem and strengthen its role as an international AI safety research hub.