Microsoft, Google, xAI sign agreements to let US government test AI models before launch — May 5, 2026, pre-deployment evaluation goes federal
Microsoft, Google, and xAI signed agreements on May 5, 2026 allowing the US government to test their AI models before public launch — the first formal pre-deployment evaluation commitments from major US labs under the new federal framework. Anthropic and OpenAI have existing equivalent agreements via the US AI Safety Institute. The federal pre-launch-testing regime is now coverage-comprehensive across the largest providers.
The procedural mechanism is the substantive piece. The Microsoft, Google, and xAI agreements are structured to give the US AI Safety Institute (US AISI) access to pre-deployment model checkpoints, with AISI running evaluations against its growing benchmark suite — including red-team probes for chemical, biological, radiological, and nuclear (CBRN) capability, agentic-autonomy thresholds, and the deception-and-deceptive-alignment evaluation patterns that have emerged from the broader alignment research community. The evaluation findings are not legally binding in the sense of granting or denying deployment authority, but they become procedurally load-bearing in the same way Anthropic's use of mech-interp in the Sonnet 4.5 safety case became procedurally load-bearing — once they exist in the safety case, they have to be addressed.
The combined effect with EU regulation is what makes the May 5 milestone consequential beyond any individual agreement. The EU AI Act's December 2, 2026 transparency deadline plus the EU AI Omnibus VII's accelerated transparency requirements create a parallel evaluation regime that the same labs have to satisfy. The US AISI agreements plus the EU AI Office's enforcement framework plus the UK AISI's third-party evaluation operations together form a tri-jurisdictional pre-deployment evaluation environment that frontier labs operate inside. The compliance cost is real, but it is the cost the labs have effectively accepted in exchange for continued market access. Expect the next round of US executive-order revisions to formalize the AISI access into something closer to binding evaluation authority rather than voluntary agreement structure.
Reuters — US AI Safety Institute pre-deployment testing agreements → · NIST — US AI Safety Institute activities → · Crescendo AI — Latest AI policy and regulation news →