Working with US CAISI and UK AISI to build more secure AI systems

Source: OpenAI Date: 2025-09-12 URL: https://openai.com/index/us-caisi-uk-aisi-ai-update

Summary

Summary

OpenAI published an update on its work with the US Center for AI Safety and Innovation (CAISI) and the UK AI Safety Institute (AISI) — describing joint research on AI security vulnerabilities, shared evaluation frameworks, and collaborative red-teaming efforts to identify and address security weaknesses in frontier AI systems.

Implications

Safety/governance thread. The CAISI/AISI collaboration is OpenAI’s most direct engagement with government AI safety evaluation bodies. Both institutes have independent evaluation access to frontier models, meaning they can assess capabilities and risks without relying on company-provided data. OpenAI’s transparency in publishing updates on this collaboration signals that the relationship is cooperative rather than adversarial — a posture choice that reflects the current US and UK regulatory environment. The September 2025 timing places this in a period when both the EU AI Act and nascent US AI governance frameworks were requiring safety evaluations, making these government-institute relationships increasingly consequential for OpenAI’s deployment approvals.