OpenAI announced on 12 September 2025 that it is collaborating with the US Center for AI Safety and International Research (CAISI) and the UK AI Safety Institute (AISI) to develop more secure AI systems [source]. The announcement describes a partnership focused on pre-deployment testing and safety evaluations for future models.

According to the blog post, OpenAI will grant both institutes early access to new models before public release, allowing independent safety assessments. The company stated that CAISI and UK AISI will conduct evaluations for risks including cybersecurity threats, biological misuse, and autonomous capabilities. OpenAI described the arrangement as part of its commitment to third-party oversight.

The post does not specify which models will be subject to these evaluations or what thresholds would prevent a model from being deployed. No technical details were provided about the evaluation methodologies or whether the institutes have authority to block releases.

OpenAI previously committed to external red-teaming in its Preparedness Framework, published in December 2023. The company has faced scrutiny over the pace of model releases and the adequacy of internal safety processes. In July 2024, several former OpenAI employees published an open letter calling for stronger whistleblower protections and independent oversight of AI labs.

The announcement follows similar commitments from Anthropic and Google DeepMind to work with government safety institutes. UK AISI was established in November 2023 following the AI Safety Summit at Bletchley Park. CAISI was announced by the US government in early 2025 as part of efforts to coordinate AI safety research across federal agencies.

No independent confirmation of the partnership's operational details has been published by CAISI or UK AISI at the time of this report.