Anthropic Seeks White House Reprieve for Mythos Models

The US Commerce Department on June 12, 2026 issued an export control directive ordering Anthropic to suspend all access to its Fable 5 and Mythos 5 models by any foreign national, whether inside or outside the United States, a category that includes non-citizen Anthropic employees. Anthropic was given approximately 90 minutes to comply and received no prior written communication of a specific national security concern, according to a source familiar with Anthropic speaking to Fortune. The company disabled both models for all customers to ensure compliance. According to Fortune and Politico, the immediate catalyst was Amazon CEO Andy Jassy, who contacted senior Trump administration officials after Amazon researchers used a series of prompts to extract cybersecurity information from the Mythos-class model that should have been restricted. White House AI adviser David Sacks stated publicly that the administration asked Amodei to fix the identified jailbreak or withdraw the model voluntarily, and that Amodei declined. Treasury Secretary Scott Bessent told Amodei directly during one call that he was making a bad decision, per Politico. Administration officials held several hours of talks with Anthropic on June 13 focused on Fable 5. By June 14, a group of experts and executives published an open letter calling on the White House to reverse the ban, and Bloomberg characterized the action as the most significant incursion ever undertaken by a government into an AI company's business. Senior Anthropic technical staff traveled to Washington over the weekend of June 14-15, per Fortune.

Anthropic complied with the legal directive while publicly disputing the rationale. In its official statement, the company said it reviewed the reported jailbreak and found it to be a narrow, non-universal bypass that 'essentially consists of asking the model to read a specific codebase and fix any software flaws' - a capability also available from OpenAI's GPT-5.5, per Anthropic. The company said it adopted a defense-in-depth strategy for Fable 5 combining guardrails with 30-day customer data retention to enable rapid jailbreak monitoring and mitigation. Anthropic stated it had never received a demonstration of a universal jailbreak - one that could broadly bypass the model's safeguards across a wide range of cyber capabilities - and argued that applying a zero-tolerance standard for non-universal jailbreaks 'would essentially halt all new model deployments for all frontier model providers.'

Reported jailbreaks in deployed models typically prompt rapid regulatory and corporate responses. Anthropic's public defense-in-depth framework distinguishes between non-universal jailbreaks - which can elicit restricted information in specific circumstances - and universal jailbreaks, which can broadly unlock restricted capabilities. The company stated that perfect jailbreak resistance is not currently possible for any model provider, a claim consistent with broadly held industry positions. Security researchers and defenders who rely on frontier models for vulnerability analysis could face access disruptions when export controls are applied, since the same prompting techniques used to simulate adversary behavior are also used in defensive testing.

The action marks the first time a US government has used export controls to suspend access to a widely deployed commercial AI model. According to Axios, a White House official said the administration does not view other models as posing the same national security risk, and that any future model crossing Mythos's capability threshold would need government review before release - signaling a potential new pre-release review requirement for frontier AI. The episode escalates a months-long standoff: earlier in 2026, the administration designated Anthropic a 'supply chain risk' for Pentagon contractors after the company declined contract terms allowing its models for 'all lawful purposes,' citing concerns over autonomous weapons. The shutdown triggered international political reaction, with European leaders including a former French prime minister and UK MPs framing it as evidence of dangerous AI dependency on US providers. Canadian Prime Minister Mark Carney warned that the action demonstrated the risk of relying on American AI infrastructure.

Track whether senior Anthropic technical staff and White House officials reach a published agreement or technical remediation plan. Watch for any formal policy statement from the Commerce Department establishing capability thresholds for pre-release government review, and whether the June 14 expert letter gains additional signatories. Monitor how other frontier model providers respond to the implied new standard for Mythos-class capabilities.