Policy Reversal on AI Safety Testing

Under the arrangement, AI models developed in the US and UK can undergo safety testing before their public release. This marks a change from the Trump administration's earlier stance, which had declined to permit such pre-launch testing of frontier AI models.

Pre-deployment safety testing is a practice under which governments or independent bodies evaluate powerful AI systems for risks — including misuse, reliability and security concerns — before those models are made available to users. The US and UK had both established AI safety institutes to carry out this type of evaluation.

Details Remain Limited

The precise terms of the testing permissions, including which models will be covered and the scope of government access, were not specified in available information. It also remains unclear whether the reversal reflects a formal agreement or an administrative policy change, and what prompted the shift in the Trump administration's position.

The move is likely to be watched closely by AI developers and policymakers in other jurisdictions, including Korea, where the Ministry of Science and ICT has been developing its own framework for AI governance and safety evaluation of large language models.

The claim that the US and UK have allowed pre-deployment safety testing of AI models, reversing the Trump administration's earlier position, is Mostly True.