A bipartisan group of four House members has given Commerce Secretary Howard Lutnick until June 26 to explain the legal foundation for the export controls his department placed on Anthropic’s two most powerful AI models. The letter, dated June 18 and signed by Reps. Sam Liccardo (D-CA), Jay Obernolte (R-CA), Scott Franklin (R-FL), and Ted Lieu (D-CA), asks Commerce to account for an action the lawmakers say “appears to represent a significant new application of export control authorities to advanced AI systems.”
The action landed on June 12, days after Fable 5 and Mythos 5 launched. The order imposed a worldwide license requirement covering exports and transfers of the models to “all foreign persons,” including foreign nationals working inside the United States. Because much of the staff at frontier labs are not U.S. citizens, Anthropic disabled both models globally rather than wall off individual users. The stated trigger was a reported method of jailbreaking Fable 5, surfaced by Amazon researchers and escalated to the administration as a national-security concern.
The letter does not litigate the merits of the jailbreak. It goes after the statute. Commerce acted, the lawmakers write, through an “is informed” letter issued under 15 C.F.R. § 744.22(b), the provision governing emerging and foundational technologies. The members pose twelve questions; four of them cut to whether the directive can stand at all.
First, they ask for the factual basis behind the finding of an “unacceptable risk” of “military intelligence end use” — the specific hook in § 744.22(a) that the control rests on. Second, they ask whether Commerce followed the interagency and public-notice process that 50 U.S.C. § 4817(b)(2)(B) requires before designating something as a controlled emerging technology, and “if not, on what basis does the Department claim authority to bypass that process.” Third, they ask whether the capability of concern is actually unique to one developer, or whether “that same or similar capability” exists in other publicly available models — including open-weight models — “that remain unrestricted and in public use.” Fourth, they ask whether Anthropic was “asked to voluntarily pause, restrict, or remediate before the formal directive was issued,” and whether it was given “an opportunity to remedy any issues.”
Those four questions track the legal critique building since the order. Analysts argue that § 744.22 and the broader Export Administration Regulations were written to control physical goods and exportable code, not access to a hosted model over an API. Alasdair Phillips-Robins, a former Commerce policy adviser, has argued publicly that querying a model through a digital interface is not an “export” as the EAR defines the term — meaning the directive, in his reading, may not restrict API or chatbot access at all. As Gizmodo put it, if Lutnick “cannot supply a clean statutory answer to Congress, the directive looks like an improvised use of peacetime trade law applied to something it was never designed to cover.”
The uniqueness question is where Anthropic’s own position does the most work. In its public statement, the company called the issue “a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws,” and said the same capability “is widely available from other models (including OpenAI’s GPT-5.5)” and “is used every day by the defenders who keep systems safe.” Anthropic added that applying that standard across the industry “would essentially halt all new model deployments.” If the capability is not specific to Fable 5, a single-vendor control is harder to defend on its face.
The lawmakers frame the stakes as precedent rather than one company’s quarter. The directive, they write, “may establish a precedent with significant implications for other developers, researchers, users, and investors throughout the AI sector,” and they want to know what “principled distinctions, if any, the Department is drawing among advanced AI models.” The letter closes by requesting a briefing and, “if necessary, a classified roundtable.”
Congress cannot compel a response, and Commerce can let the date pass. But a non-answer to a bipartisan oversight letter — on a control whose statutory basis outside analysts are openly questioning — is its own data point. What to watch is narrow: whether Commerce supplies a clean statutory citation and a factual basis by June 26, or whether the deadline passes with the models still dark and the legal question unanswered.