Anthropic Takes Crisis to Washington as Fable 5 and Mythos 5 Remain Offline Worldwide
Senior executives from the company met with the White House and Department of Commerce on Monday to try to reverse the order that has kept Fable 5 and Mythos 5 offline since the night of June 12.
On Monday, Anthropic sent its senior technical team to Washington for meetings with the White House and Department of Commerce in an attempt to reverse the executive order keeping the Fable 5 and Mythos 5 models offline globally since the night of June 12. The meetings lasted for hours and are expected to extend into Tuesday, according to reporting from Fortune. No public resolution had been reached by the time this article was published.
The crisis began three days after the launch on June 9 of Fable 5 as a public model and Mythos 5 as a tool restricted to qualified cybersecurity defenders. Researchers at Amazon, Anthropic's largest investor, identified a 'fix this code' type jailbreak that was capable of extracting dangerous outputs from both models. Amazon CEO Andy Jassy escalated the finding directly to Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross, according to Fortune.
Lutnick's Letter and Amodei's Refusal
On the night of June 12, Lutnick sent a letter to Anthropic CEO Dario Amodei, informing him that Fable 5 and Mythos 5 would now be subject to export controls. The order requires Anthropic to block access to the models for any foreign citizen, anywhere in the world, including the company's own employees. By midnight, both models had been taken offline globally.
According to Anthropic, the letter did not contain a specific justification related to national security. David Sacks, AI advisor to the Trump administration, claims that the administration offered Amodei the choice between fixing the jailbreak or taking the models offline. Sacks states that Amodei refused. Anthropic contests this version. Lutnick, in a public statement, cited an 'unacceptable risk' of the models being 'diverted to military intelligence users in China, Russia, or other countries of concern.' The company asserts that the issue of Chinese access was never raised in the initial conversations about the jailbreak.
On Saturday, Tom Brown, co-founder and chief compute officer of Anthropic, and Sarah Heck, head of public policy, spent hours in calls with Lutnick and Cairncross. On Sunday, the company dispatched a technical team to Washington. The agenda for Monday's meetings combines three fronts: security protocols that satisfy national defense concerns, a framework for international distribution of the models, and terms for federal agencies to resume or expand their use of the technology.
The Impact on the UK, Japan, and Brazil
The impact of this order is not limited to the United States. It affects any Anthropic client outside the U.S. In the UK, Germany, and Japan, IT teams spent the weekend migrating workloads to alternatives such as Claude 4.5, GPT-5, and Gemini 3 Pro. Cybersecurity firms consulted by industry media reported a decline in the quality of binary analysis and exploit chain detection, two use cases for which Mythos 5 had been positioned as a benchmark.
In Japan, contracts for Claude in banking compliance and back-office automation that were set to go live in 2025 are now in unforeseen force majeure territory. The typical SLA clauses in AI SaaS contracts for that year did not account for midnight export orders as a discontinuity scenario. In the UK, part of the public sector's AI effort relies on Anthropic's models, and the Department of Business and Trade is monitoring the case through the lens of digital sovereignty, a debate that had taken a backseat until last week.
For Brazil, the effect is lateral but instructive. Major Brazilian banks maintain a diversified portfolio of American models in pilots, and the abrupt discontinuity due to an American executive order is exactly the scenario outlined in internal AI risk manuals throughout 2026. The next review round is likely to mandate the adoption of at least two competing models for any regulated workload, with explicit clauses addressing geopolitical supplier risk.
The Cost of AI Monoculture
The Anthropic case demonstrates in one week what two years of risk reporting have not achieved. It shows the global C-level executives that the generative AI stack is a supply chain with a single point of failure. A letter delivered to a single CEO in San Francisco has taken offline capabilities that enterprise clients around the world had contracted.
Anthropic is heading to Washington to negotiate the return of the models. Even if successful, and there are indications that discussions are progressing, according to Fortune, the precedent is set. For corporate buyers, the question to be answered in the next procurement is no longer which model to choose, but how many suppliers to keep active so that a midnight executive order in another country does not disable a critical production service.