The US frontier-model order reviews without licensing
EO 14409 adds voluntary cybersecurity mandates and a pre-release review framework for frontier models — and states explicitly that it creates no licensing or preclearance requirement. The distinction is the entire American position.
The June executive order does two things that sound contradictory until you read the carve-out. It establishes a pre-release review framework for models designated "frontier", and it explicitly creates no licensing regime and no preclearance requirement.
That is not a drafting hedge. It is the operative design. A review you must submit to before shipping is a licence by another name; a review that runs without the power to withhold approval is a disclosure exercise. The order chose the second, and said so in terms, presumably to foreclose the argument that it had quietly chosen the first.
Put it beside the EU and the divergence is now structural rather than rhetorical. Brussels has a statute, an office, and — as of this week — a demonstrated path from finding to penalty. Washington has a framework whose obligations are voluntary and whose review carries no gate.
This follows the revocation of EO 14110 on the administration's first day back, which traded mandatory reporting thresholds and civil-rights enforcement for a stated policy of removing barriers. EO 14409 is the constructive half of that trade: something rather than nothing, structured so that it cannot become a bottleneck.
For anyone shipping into both jurisdictions the practical consequence is that the EU sets the floor. Voluntary US obligations do not relieve you of the Act, and the Act now has receipts.
Collibra — AI regulatory compliance in 2026: EU AI Act, US orders, and state laws → · Gunderson Dettmer — 2026 AI Laws Update: Key Regulations and Practical Guidance → · Smith Anderson — The Future of AI Compliance — Preparing for New Global and State Laws →