US Government Blocks Anthropic's Fable 5 and Mythos 5

Anthropic received a US export control directive that required it to suspend access to Fable 5 and Mythos 5 for "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees," the company wrote in a public statement. The company added, "We received the directive from the government today at **5:21pm (ET)," and said, "The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance." NBC News, citing an administration official, reports the letter was sent from Commerce Secretary Howard Lutnick and prepared with help from the Commerce Department's Bureau of Industry and Security. Axios reported the directive was first communicated verbally at 1 p.m. ET Friday, giving Anthropic roughly 90 minutes to act, followed by a written letter around 5:30 p.m.; users lost access by approximately 10 p.m.

Per Axios reporting, Amazon called senior administration officials Thursday night to share a report showing how researchers had jailbroken Anthropic's Mythos model and accessed cybersecurity-related capabilities that were supposed to be restricted. At least five other companies also contacted senior officials Thursday evening and Friday morning. Anthropic had previously notified the government multiple times about the planned June 9 release of Fable 5 - described as a guardrailed general-use version of Mythos - and the government did not object at that time, according to a source close to the company (Axios).

Public reporting and Anthropic's statement highlight the government concern as centered on a claimed jailbreak or bypass of Fable 5. Anthropic's statement says, "Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5," and that a demonstrated technique identified a small number of previously known, minor vulnerabilities. Anthropic also wrote that those vulnerabilities "appear relatively simple, and we have found that other publicly-available models are able to discover them as well." Luta Security CEO Katie Moussouris, who Anthropic shared the Amazon report with, told Axios: "The government's response seems way out of line with what's actually in the research report."

Industry-pattern observations: Advanced models are commonly layered with external or independent safety classifiers and red-team programs rather than relying solely on a single model checkpoint. Reporting indicates Anthropic engaged with multiple external partners and government teams in pre-launch red-teaming; the company says it worked with the US government, the UK AISI, and private third-party organisations to test safeguards for thousands of hours.

Mythos 5 had been available only under restricted arrangements, reported TechCrunch and CNBC, as part of a cybersecurity program called Project Glasswing that Anthropic described as limited to a small set of vetted defenders. Launch partners for Project Glasswing, per SiliconANGLE, include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Fable 5, released publicly days earlier, was presented by Anthropic as a guardrailed version of Mythos intended for broader use; multiple outlets noted benchmark results showing Fable 5 among the most capable publicly available models. The Commerce Department action, as reported, is the first publicly known instance of federal export-control intervention that forced a leading AI lab to take a deployed model offline worldwide to meet a directive aimed at foreign-national access.

The incident combines technical risk (jailbreaks and misuse potential), regulatory reach (export-control tooling applied to models), and disclosure dynamics (what researchers show government vs public). An administration official told Axios they do not view other models as national security threats "because they do not surpass the bar that Mythos set," and that anything at Mythos level or above would need to go through the administration for national security review. One person familiar described the situation to Axios as "a de-facto licensing regime."

• •Whether the Commerce Department or other agencies provide a written rationale or technical evidence beyond the verbal briefing described in reporting.
• •If Anthropic or other labs disclose remediation steps and whether those steps are accepted by regulators.
• •How enterprise customers and Project Glasswing partners respond to the sudden unavailability of Mythos 5 for defensive workflows.

For practitioners: Monitor vendor risk clauses, contractual SLAs for model availability, and internal threat modeling for jailbreak vectors. Also track whether this prompts changes to red-team disclosure practices or creates new compliance workflows for foreign-national access to sensitive model capabilities.

This is a high-profile example of export-control authorities intersecting with model safety and release practices. It is both a technical incident about jailbreak demonstrations and an early test case for how US national security tools can shape the operational availability of advanced AI models.