Andy Jassy's Offhand Call Triggered Anthropic's Model Shutdown

The Friday evening that Anthropic's Fable 5 and Mythos 5 models went offline reportedly began not with a formal government investigation but with an offhand mention in an otherwise routine phone call. According to Fortune's reporting, Amazon CEO Andy Jassy raised a jailbreak vulnerability his researchers had found in Fable 5 during an unscheduled call with White House officials on June 11. Those officials directed him to notify Treasury Secretary Scott Bessent, and that referral set off a chain of events ending with Commerce Secretary Howard Lutnick issuing a 90-minute ultimatum: fix the jailbreak or pull the models.

By 10 p.m. on June 13, Anthropic complied. The shutdown was framed around export control restrictions, meaning the government's lever was not a product recall in the conventional sense but a trade enforcement mechanism applied to a domestic AI system. Export controls were designed for hardware and classified technology crossing borders, not for pulling a live API used by commercial customers. Their application here suggests the administration is willing to improvise regulatory tools when existing ones do not quite fit.

The industry pushback was pointed. Danny Jenkins, CEO of ThreatLocker, reportedly warned that export controls risk disadvantaging U.S. AI companies, arguing that "criminals don't care whether they're allowed to export something," since restricting domestic models does not remove the underlying vulnerability from the global threat landscape. That critique carries weight: a jailbreak narrow enough that Anthropic judged insufficient to justify recalling a model does not disappear when the model goes offline for American users.

What the reporting does not give you is clarity on what legal authority Lutnick actually invoked, or whether Anthropic has a concrete path to restoring the models. As of Thursday evening, negotiations were reportedly ongoing but resolution details remained unclear. Take the specifics as reported rather than settled; fast-moving regulatory standoffs tend to produce accounts that shade toward one side's preferred narrative, and this one reportedly followed months of tensions between Anthropic and the Trump administration over Pentagon contracts and AI regulation advocacy.

The forward-looking question is who builds the institutional capacity to navigate this without a 90-minute clock. Companies with mature safety-disclosure programs and active government relations now have a structural advantage that will not show up in any model benchmark. If the ongoing negotiations produce a durable framework, the irony is that Anthropic and the larger labs may benefit most, gaining codified rules that turn the next jailbreak discovery into an orderly procedure rather than an emergency shutdown.