US Restores UK AI Safety Institute Access to Frontier Models for Testing
The United States has reinstated the UK AISI's ability to review advanced AI models from major tech firms, reversing a recent directive that halted data sharing.
Key points
- The US government restored access for the UK AI Safety Institute to review frontier AI models.
- This move reverses a previous order given to OpenAI and Anthropic to stop sharing model data.
- The prior restriction also prevented regulators from reviewing models from Google DeepMind.
The United States has restored the UK AI Safety Institute’s access to the latest frontier AI models for safety testing. This development reverses a directive issued last month that ordered major artificial intelligence developers to halt the sharing of their proprietary model data with British regulators.
According to Tony Diver at the Telegraph, the US government had previously blocked regulators from reviewing frontier models developed by Anthropic, OpenAI, and Google DeepMind. The recent change allows the UK AISI to resume its evaluation of these advanced systems, ending the temporary suspension of data access that had been in place.
Regulatory Shift and Model Access
The reversal affects the ability of the UK AI Safety Institute to conduct safety assessments on the most advanced AI systems currently in development. The US had explicitly ordered OpenAI and Anthropic to stop sharing their models with the institute. This restriction effectively cut off the British regulatory body from critical data needed for its safety testing protocols.
The blockage extended beyond just those two companies. The US also prevented the review of frontier models from Google DeepMind. This comprehensive restriction halted the UK AISI’s access to a significant portion of the leading AI technologies available for safety evaluation at the time of the order.
Impact on Safety Evaluations
The UK AI Safety Institute relies on access to these frontier models to perform rigorous safety testing and risk assessments. The previous US directive created a barrier that prevented the institute from reviewing the specific architectures and behaviors of these advanced systems. Restoring access allows the institute to continue its mandate of evaluating AI safety risks.
Tony Diver’s reporting indicates that the US had actively blocked the regulators from this review process. The reinstatement of access suggests a shift in the US position regarding international cooperation on AI safety testing, at least for these specific entities and models. The UK AISI can now proceed with its safety evaluations using the latest model versions.
What should you do?
Organizations relying on AI safety certifications or risk assessments from the UK AISI should monitor official announcements regarding the scope of resumed testing. Professionals in AI governance should track how this restored access influences future regulatory requirements for model developers. Companies should remain aware that access to frontier models for safety testing is subject to international policy changes.
What to do and how to stay safe: UK AI Safety Institute
- Monitor official communications from the UK AI Safety Institute for updates on resumed safety testing schedules and methodologies.
- Review internal compliance protocols to ensure alignment with any new requirements resulting from restored regulator access to frontier models.
- Track policy developments regarding international data sharing agreements between US AI developers and foreign safety institutes.
- Assess how changes in regulatory access might impact the timeline for safety certifications of advanced AI systems in your portfolio.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Which companies were ordered to stop sharing models with the UK AISI?
The US ordered OpenAI and Anthropic to stop sharing models, and also blocked reviews of models from Google DeepMind.
Who reported that the US restored access to the UK AI Safety Institute?
Tony Diver at the Telegraph reported that the US restored the UK AISI's access to the latest US models for safety testing.
What was the previous status of the UK AISI's access to these models?
The US had blocked regulators from reviewing frontier models from Anthropic, OpenAI, and Google DeepMind last month.
