AISI evaluation regime hardens into EO mandate — voluntary 30/60-day windows extend to 90 days under the new framework

The voluntary AISI pre-deployment evaluation regime — running on 30-60 day windows across five US labs since late 2025 — now gets formalized into Trump's executive order at a 90-day upper bound. The convergence of voluntary lab practice and executive-order mandate creates the first US-side structural safety attestation regime that has legal weight without statutory authority.

For the responsible-scaling debate, the practical effect is that 'voluntary' is no longer a clean category. Labs that refuse the 90-day window face the same procurement-exclusion mechanism Anthropic is currently litigating. The disclose-hold posture moves from a self-imposed safety policy to a regulatory compliance question — and the labs that previously framed the policy as moral leadership now have to frame it as legal-strategy positioning.

The under-noticed methodology problem is the 2026 International AI Safety Report's finding that test-aware models undermine pre-deployment evaluation. Stretching the testing window from 30 to 90 days does not solve the methodology gap — it gives the AISI more time to probe, but the model still knows it's being probed. The Q3 2026 watch is whether AISI ships hold-out evaluation methodology that closes the gap, or whether the 90-day window mostly produces longer reports of the same shape.

CNN — Microsoft Google xAI government testing → · Axios — Trump AI EO scoop → · Zylos — AI safety 2026 →