U.S. Frontier-Model Review Stalls; EU Sets Timelines
Yesterday's clearest U.S. AI-governance development was a pause. President Trump postponed a planned executive order that would have created a voluntary process for federal pre-release review of frontier AI models, including early access for government testing of security and national-security risks. For companies, that leaves immediate obligations unchanged; for Washington, it reflects an unresolved policy dispute over how much scrutiny advanced models should face before launch.
Europe moved more concretely. EU lawmakers agreed to ban AI nudifier tools used to generate child sexual abuse material and nonconsensual intimate imagery, with a Dec. 2 compliance deadline for providers. In the same package, they pushed some AI Act duties for high-risk systems into 2027 and 2028 because the underlying technical standards are not ready.
In financial regulation, Bank of England and Financial Stability Board officials sharpened the message that a human checking individual generative-AI outputs is not, by itself, an adequate control. For firms using AI in workflows such as trading or advice, the expectation is moving toward responsibility for the full system and process. That is not a new rule on its own, but it is a clear supervisory direction of travel.
Key Points
- The White House delayed a planned executive order on voluntary pre-release review of frontier AI models, leaving the proposal unresolved.
- The draft U.S. framework would have enabled federal testing of advanced models before public release, with reported review windows of up to 90 days.
- EU lawmakers agreed to ban AI nudifier tools tied to child sexual abuse material and nonconsensual intimate imagery, with a Dec. 2 compliance deadline.
- Some EU AI Act obligations for high-risk systems were deferred to 2027 and 2028 because technical standards are not yet ready.
- Bank of England and Financial Stability Board officials said human review alone is not a sufficient control for generative AI in regulated workflows.
Implications
U.S. frontier-model oversight remains in a voluntary and politically unsettled phase rather than moving into a stable federal review regime.
EU compliance planning now needs two tracks: near-term attention to prohibited-use enforcement and a longer runway for some high-risk AI Act obligations.
Regulated firms should expect supervisors to look beyond output review toward controls over the full AI workflow, including responsibility, monitoring, escalation, and auditability.
Things to watch
Watch
Whether the White House revives the frontier-model review order, narrows it, or drops the pre-release access concept altogether.
Watch
How quickly the EU turns the nudifier ban into operational enforcement and when the delayed high-risk standards are finalized.
Watch
Whether financial supervisors turn the view that human-in-the-loop is insufficient into formal guidance or exam expectations.