President Donald Trump on June 2, 2026 signed an executive order laying the foundation for federal testing of the world's most powerful AI systems before they are publicly released. The order asks AI companies to voluntarily submit frontier models for government cyber-capability evaluation up to 30 days before launch, and directs agencies to stand up an 'AI cybersecurity clearinghouse' to share vulnerability information.
Voluntary, not mandatory
The order explicitly bars the federal government from creating a mandatory licensing or preclearance requirement for new AI models. Pre-release testing is a request, not a rule — designed to align with industry capacity rather than gate releases. The 30-day window replaces an earlier draft that proposed up to 90 days; tech executives and some White House advisers pushed back that the longer window would slow shipping cycles and disadvantage US labs against Chinese frontier developers.
Who's expected to participate
The testing program leans on voluntary collaboration from Anthropic, OpenAI and Google, the three labs that account for most US frontier model releases. All three already publish system cards and run pre-deployment red-teaming; the order codifies a federal layer on top of that work rather than replacing it. Anthropic, which is now in the public-markets pipeline via a confidential S-1 filing, will face additional scrutiny as a public-company candidate.
Cybersecurity clearinghouse and benchmarks
The order directs federal agencies to develop benchmarks that assess AI models' cyber capabilities — including vulnerability discovery, exploit generation and autonomous offensive operations — and creates an AI cybersecurity clearinghouse to consolidate and share findings across government, infrastructure operators and labs. The clearinghouse is intended to give critical-infrastructure operators a faster path to learn about model-driven cyber risks already visible to the labs.
What it doesn't do
The order does not address content moderation, copyright training data or competition. It also leaves untouched the broader AI capex push by hyperscalers, including Alphabet's $80 billion equity raise for AI infrastructure announced 24 hours earlier. By stopping at voluntary review and a cyber-information clearinghouse, the White House signaled it does not want to constrain the buildout side of US AI policy in this term.
Industry reaction
Industry response was muted-to-positive: cybersecurity-focused groups welcomed the clearinghouse; some civil-society groups argued voluntary review without enforcement is too weak; and frontier labs publicly committed to participate. The narrower scope keeps the EO more compatible with state-level efforts and existing NIST AI Risk Management Framework guidance.
Reporting based on coverage from NPR, NBC News, CNBC, Cybersecurity Dive, Roll Call and Axios.