Anthropic Sends Technical Team to Washington to Attempt Reactivation of Mythos 5 and Fable 5

Anthropic has sent part of its senior technical team to Washington to discuss with the White House the reactivation of Claude Mythos 5 and Claude Fable 5, which have been offline globally since the evening of Friday (12). The information was initially reported by Axios on Sunday (14), and the White House, in a statement, described the first meeting as "productive and constructive."

The initial meeting brought together CEO Dario Amodei, White House Chief of Staff Susie Wiles, and Treasury Secretary Scott Bessent. According to the official note released by the White House, the agenda involved "opportunities for collaboration in cybersecurity, AI safety, and U.S. leadership in the race for artificial intelligence." The meeting did not include Secretary of Commerce Howard Lutnick, who signed the export control order, nor Secretary of Defense Pete Hegseth, with whom Anthropic has been in an ongoing legal dispute since March.

The Letter that Took the Models Offline

Lutnick's directive reached Amodei on Thursday (11) at 5:21 PM Washington time, according to Axios. The document subjects Mythos 5 and Fable 5 to export control regulations and requires a license for any access by foreigners, inside or outside the United States. As Anthropic is unable to separate foreign users from American users in real-time within its commercial product, the company turned off both models worldwide around midnight from Friday to Saturday. Only Claude Opus 4.8 and previous generations remain available.

The order was preceded by a warning from Amazon. AWS researchers were able, with a specific sequence of prompts, to extract restricted information from Mythos regarding cyber-attack techniques. Andy Jassy, CEO of Amazon, brought these findings to senior officials in the administration on Thursday. Anthropic publicly maintains that it is a narrow bypass based on asking the model to read a specific codebase, rather than a universal jailbreak of protections.

The Interrupted Commercial Window

Fable 5 was released on June 9 with a free offer until June 22 for Pro, Max, Team, and Enterprise subscribers. Mythos 5, focused on cybersecurity, had been distributed since June 2 to 150 organizations in over 15 countries. Anthropic announced refunds for those who paid for access, without disclosing the total amount. The company's annualized revenue reached $47 billion in May, according to Daniela Amodei, the company's president, in a panel at Bloomberg Tech on June 4.

Salesforce, which confirmed in May its commitment to spend $300 million on Anthropic tokens only in 2026, lost access to the vendor's newest models. Marc Benioff, CEO of Salesforce, justified the expenditure in the All-In podcast by saying that "everything is going to be cheaper to do" thanks to the use of AI for coding. For integrators like Capgemini, Infosys, and CI&T, who had been selling Claude-based deliverables to European and Brazilian banks, the interruption forced them to trigger contingency plans with Gemini and GPT amid deployment sprints.

The Pentagon at the Core of the Dispute

The current dispute carries over from previous tensions. On February 27, Trump ordered federal civilian agencies to cease using Anthropic's technology. In March, the Department of Defense labeled the company as a supply chain risk after it refused to release its models for use in autonomous weapons and domestic surveillance. Anthropic sued the government that same month and obtained a partial injunction in a federal court in California, a decision that has yet to be reviewed by a higher court. It was the first American company to receive the label, typically reserved for suppliers linked to China, Russia, and Iran.

Bessent was chosen as the interlocutor by Anthropic precisely because he is outside the Pentagon's hardline faction. The Treasury Secretary coordinates the government's broader commercial strategy and has direct concerns about the impact of the measure on the U.S. position in the global AI market. Wiles, according to Axios, saw the discussions as an opportunity to avoid a prolonged fight that would serve as ammunition for the European and Chinese narrative that the U.S. is using AI as a commercial weapon.

The Bottom Line for Corporate Buyers

Regardless of how the impasse is resolved, there remains an operational lesson for CIOs and CISOs in any market: a SaaS model can be turned off by executive decision in the United States without notice to foreign buyers. The relevant contract is not only the vendor's but also the jurisdiction in which it operates. Corporate risk teams in Germany are already citing the shutdown to support BaFin's requirements on exit plans from AI providers. Banks in Japan and fintechs in Brazil that standardized critical workflows on Claude are entering mandatory architectural reviews in the coming weeks.