Monument Advocacy Principal and Chief AI Officer Joseph Hoefer examines a First Amendment question raised by Trump’s June 2 AI executive order: can the government require permission before an AI model is published?
President Trump’s June 2 executive order on AI and cybersecurity explicitly disclaims any authority to require government approval before a new AI model, including a “frontier model,” is published or distributed. In a new Bloomberg Law piece, Joseph Hoefer argues that disclaimer may be constitutionally significant, and previews the legal fight likely to follow.
Hoefer traces the question back to 1990s court battles over encryption export controls (Bernstein v. DOJ, Junger v. Daley), which asked whether source code is protected speech even though it also performs a function. He argues AI model weights present a harder case: unlike human-written code, weights are numerical parameters largely unreadable to people, which could shape how courts eventually weigh a developer’s right to publish against the government’s interest in restricting AI capabilities.
The piece draws a key distinction for companies and policymakers: regulating how an AI model is used is different, legally, from requiring permission before it’s published. Hoefer points to Defense Distributed v. State Department and the pending Anthropic PBC v. U.S. Department of War as early signals of how that line may get tested in court.
His guidance for companies with open-weight release programs: document the research purpose behind releases, separate deployment policy from publication policy, and watch closely how new AI legislation is structured, since reporting rules, waiting periods, and approval gates can raise very different constitutional questions.