America Needs an Off-Ramp Between Doing Nothing and Shutting AI Down

In mid‑June 2026 the U.S. Commerce Department ordered Anthropic to shut down its Fable 5 and Mythos 5 models worldwide because of alleged national‑security concerns, forcing the company to suspend the models for 18 days. The decision came after Amazon security researchers reportedly found a "jailbreak" and the order was issued without detailed justification, exposing the absence of a transparent, standardized review of frontier AI.

The episode shows that the U.S. lacks a trusted mechanism to assess security risks of cutting‑edge models and to apply proportionate remedies. The author argues for a statutory AI security review agency with independent legal authority, fixed leadership terms, protected funding, and a tiered review process that includes a 30‑day pre‑release assessment, a 72‑hour emergency review, and written determinations subject to judicial review. Such a body would replace ad‑hoc export controls and opaque pressure tactics, reduce the risk of industry capture, and provide clear, predictable outcomes for developers and users.

A transparent U.S. review regime would also restore confidence among allies who view current export controls as unpredictable and weaponized. By offering mutual recognition of security evaluations and shared testing standards, the agency could prevent a shift toward Chinese AI stacks and preserve American leadership in AI innovation and diffusion. The article concludes that without a statutory agency, the U.S. risks ceding competitive advantage to China and undermining global AI governance.

(Source：War On The Rocks)