The ban lasted approximately 2.5 weeks, beginning around June 12 when Anthropic commenced coordination with the government. The directive invoked U.S. export control laws and required licensing for any transfer of these models to foreign persons. Anthropic has since deployed new safety classifiers designed to block the specific bypass technique in over 99 percent of cases and is collaborating with the government on a broader AI jailbreak response framework under Project Glasswing. The company maintains that the vulnerability was narrow and that comparable capabilities exist in other publicly available models, though the government has not released detailed public evidence beyond its national security assertion.
Attorneys should monitor this as a significant reversal of rare government restrictions on frontier AI technology. The swift restoration suggests the Trump administration may be recalibrating its approach to balance AI innovation against national security concerns. For practitioners advising AI companies, this episode underscores the regulatory risk posed by safety vulnerabilities and the potential for sudden, disruptive government intervention. The outcome also demonstrates that proactive safety measures and government coordination can facilitate rapid resolution of export control disputes in the AI sector.