The Commerce Department and Anthropic held an emergency meeting Monday (June 15) over the government’s recent export controls on the company’s latest AI model, but the talks ended with no deal — Claude Fable 5 remains under a global ban. The administration’s order was driven by security concerns: fear that if Fable 5’s safety guardrails were jailbroken, the model could degrade and unleash Mythos-level capabilities for network attacks and vulnerability discovery, which foreign military intelligence agencies could then exploit.
The closed-door session in Washington was led by Anthropic co-founder and chief compute officer Tom Brown, frontier red team lead Logan Graham, and senior safety researcher Nicholas Carlini. Officials from the Office of the National Cyber Director and the Commerce Department listened in. The team argued that the government’s jailbreak fears were overblown, but Commerce officials and researchers from the AI Standards and Innovation Center (CAISI) weren’t convinced. The government side insisted that the model’s safety protections could be bypassed and demanded Anthropic fully address the jailbreak vulnerabilities.
Both sides are now urgently working on next steps. Commerce Secretary Howard Lutnick joined the meeting remotely from the G7 summit in Évian-les-Bains, France, and has been in regular contact with Anthropic CEO Dario Amodei. Since both are in France for the summit, Lutnick and Amodei are expected to meet face-to-face on the sidelines this week. In support of Anthropic, more than 80 cybersecurity professionals — including executives from Nvidia, Adobe, and Zoom, as well as academics — signed an open letter on June 14 arguing that the ban not only hurts defenders but also undermines U.S. leadership in AI.