Grant the AI Security Institute (AISI) legal authority and resources to evaluate AI models before their deployment in the UK market and influence their regulation. Establish AISI sub-teams within regulators to effectively oversee and manage the sector-specific regulation of AI models.

By enshrining AISI’s mandate, powers and collaboration with regulators, we think the risks of advanced AI systems could be mitigated, safeguarding the public, and promoting ethical AI development. We think the AISI should have the legal authority to evaluate new AI models before public release and block those that pose a threat. We think this approach balances innovation with safety while leveraging existing expertise. We think empowering the AISI would also protect its impartiality and longevity, ensuring consistent and effective AI oversight.