OpenAI, the company that arguably started the generative AI arms race, is now actively participating in government-led safety testing of its most advanced models. The move is part of a broader industry shift toward voluntary cooperation with federal regulators, one that has quietly expanded to include most of the major players in frontier AI development.
From handshake deals to structured agreements
The formal groundwork was laid on August 29, 2024, when OpenAI and Anthropic each signed a memorandum of understanding with the US AI Safety Institute. The agreement gave government researchers collaborative access to frontier models, allowing them to evaluate capabilities, risks, and potential mitigations both before and after public release.
By August 2025, the two companies publicly released findings from a joint evaluation of each other’s models, specifically focused on misalignment issues.
By May 2026, the US expanded its voluntary safety pacts to include Google, Microsoft, and xAI. Then, on June 2, 2026, the White House issued a directive establishing benchmarks for what it calls “covered frontier models” and creating voluntary frameworks that would give the government up to 30 days of pre-release access.
