The U.S. Requires a Middle Ground Between Inaction and Completely Banning AI Systems
Summary
In June 2026, the U.S. Commerce Department ordered Anthropic to restrict access to its two leading AI models, Fable 5 and Mythos 5, for 18 days following reports from Amazon security researchers about potential vulnerabilities in the models' safety guardrails. The shutdown exposed a critical gap in U.S. policy: there is no standardized, transparent process for evaluating national security risks posed by frontier AI models or for applying proportionate responses when threats are identified. The episode also occurred against a backdrop of political tension, with the Pentagon having previously labeled Anthropic a supply chain risk and senior officials publicly attacking the company's leadership, raising questions about whether the action was driven by legitimate security concerns or personal animosity. The legal foundation for the government's action is also contested, as scholars argue that existing export control laws may not apply to foreign users accessing AI models hosted on American servers, and a legal challenge has already been filed invoking the major questions doctrine. The author argues that Congress must establish a dedicated AI security review agency with purpose-built legal authority, strict procedural safeguards, proportionality requirements, and judicial oversight to prevent both government overreach and industry capture.
Key Takeaways
- 1. The U.S. government shut down two of Anthropic's top AI models for 18 days in June 2026 using emergency export controls, revealing the absence of a formal AI security review process
- 2. The action lacked transparency, as the Commerce Department's letter to Anthropic did not specify the exact national security concerns prompting the restriction
- 3. Political tensions between the Pentagon and Anthropic raise legitimate questions about whether the shutdown was motivated by security evidence or institutional retaliation
- 4. The legal authority used to justify the shutdown is legally questionable, with critics arguing existing export control laws do not clearly apply to cloud-hosted AI models accessed by foreign users
- 5. The author calls on Congress to create a dedicated AI security review agency paired with narrowly tailored emergency authority, judicial review, and anti-capture safeguards to fill the current institutional void