The US Commerce Department partially lifted its restriction on Anthropic's frontier models on 27 June, turning a hard cut-off into a fast-moving one. The reversal came fifteen days after the original order that had pulled the two models worldwide, converting an outright block into a partial, conditional restriction whose remaining scope is still unsettled.
Commerce did not spell out which users regain access or on what terms. The speed of the climbdown, barely two weeks from order to partial lift, shows the restriction working as a policy lever rather than a fixed technical limit, one Washington can tighten or loosen at will. The same discretion runs through the 24 July trade ruling now approaching.
The episode lands five weeks before the EU AI Act's general-purpose AI (GPAI) rules start to bite on 2 August, the date the bloc's AI Office can begin fining model providers. European enterprises weighing procurement against that clock now face a moving target. A model whose availability can change inside a single quarter is hard to standardise on, and that hands the European labs an opening they did not have to win.
