Anthropic suspends Fable 5 and Mythos 5 for foreign nationals after US directive
New evidence says the order applies to foreign nationals, including Anthropic employees, and centers on a narrow jailbreak tied to fixing flaws in a codebase. Other Anthropic models stay online, so this is a targeted access suspension rather than a full shutdown.
TL;DR
- In Anthropic's directive statement, the company said the US government ordered it to suspend Fable 5 and Mythos 5 for any foreign national, including foreign national employees, and that the compliance effect was to disable both models for everyone; directive statement excerpt captures the same scope.
- According to Anthropic's statement, the trigger was a suspected jailbreak tied to asking Fable 5 to read a specific codebase and fix software flaws, but Anthropic said the demonstrated findings were minor, previously known vulnerabilities that other public models could also find; Simon Willison's note excerpted the same passage.
- Anthropic's June 9 launch post had positioned Fable 5 as a public Mythos-class model with conservative classifiers that routed risky prompts to Opus 4.8 in under 5% of sessions on average, while Mythos 5 exposed the same base model with some safeguards lifted for Project Glasswing partners; launch post excerpt summarizes those mechanics.
- In the main HN thread, commenters quickly treated the order as more than a product outage: discussion highlights framed it as a possible precedent for API-level export controls and a reminder of how much operational risk sits inside cloud-only model dependencies.
You can read Anthropic's full directive statement, compare it with the earlier launch post, and skim Simon Willison's link post for the cleanest excerpt of the foreign-national language. The weirdest detail is that Anthropic says the government only provided verbal evidence of a narrow jailbreak, while the public effect was a full shutdown of both models.
Foreign nationals
Anthropic Suspends Fable 5 and Mythos 5 Access Following US Government Directive
Anthropic has suspended all customer access to its Claude Fable 5 and Claude Mythos 5 models following a legal directive from the US government. The directive cites national security concerns regarding a potential narrow jailbreak that could allow the models to identify and fix software vulnerabilities. While Anthropic is complying with the order, the company expressed disagreement with the government's decision, characterizing the jailbreak risk as minimal and noting that similar capabilities are available in other industry models. Anthropic is currently working to resolve the situation and restore access.
Anthropic said the order applied to any foreign national, whether inside or outside the US, including Anthropic employees who are foreign nationals. The company also said the practical consequence was broader than that wording, because it had to turn off Fable 5 and Mythos 5 for all customers to stay compliant.
That makes this a targeted model shutdown, not a platform-wide one. Anthropic's statement explicitly said all other Anthropic models stayed online.
The jailbreak claim
Statement on the US government directive to suspend access to Fable 5 and Mythos 5
Statement on the US government directive to suspend access to Fable 5 and Mythos 5 Well this is nuts: The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected. We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or "jailbreaking" Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass. [...] To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws. Our understanding is that one potential jailbreak was shared with the government. We have reviewed the report and validated that the level of capability displayed
In the full Anthropic statement, the company said the government did not provide specific written details beyond national security concerns. Anthropic's description of the issue is unusually narrow: a technique for getting Fable 5 to read a specific codebase and identify or fix software flaws.
Anthropic also said the demonstration it reviewed surfaced only a small number of previously known, minor vulnerabilities. Later in the same statement, it added that the capability level in the report was already available from other models, including GPT-5.5.
Safeguards and fallback routing
Anthropic Suspends Access to New Claude Fable 5 and Mythos 5 Models
Anthropic has suspended access to its newly launched models, Claude Fable 5 and Claude Mythos 5, and is working to restore services. Claude Fable 5 was released for general use, while Claude Mythos 5 was initially restricted to Project Glasswing partners for cybersecurity applications, featuring lifted safeguards. Both models were priced at $10 per million input tokens and $50 per million output tokens.
The June 9 launch post matters because it described Fable 5 as a Mythos-class model made safe for general use by wrapping it in classifiers. Prompts in cybersecurity, biology, chemistry, or distillation could be diverted to Claude Opus 4.8 instead, and Anthropic said those safeguards triggered in less than 5% of sessions on average.
Anthropic used the suspension statement to defend that design in more detail. It said no red team had found a universal jailbreak, that non-universal jailbreaks are likely unavoidable across providers, and that Fable 5 was launched with defense in depth: strong safeguards, monitoring, and a 30-day retention policy for customer data.
Mythos 5 and Project Glasswing
Anthropic Suspends Access to New Claude Fable 5 and Mythos 5 Models
Anthropic has suspended access to its newly launched models, Claude Fable 5 and Claude Mythos 5, and is working to restore services. Claude Fable 5 was released for general use, while Claude Mythos 5 was initially restricted to Project Glasswing partners for cybersecurity applications, featuring lifted safeguards. Both models were priced at $10 per million input tokens and $50 per million output tokens.
Before the shutdown, Anthropic had split the release into two surfaces:
- Fable 5: the public version of the Mythos-class model.
- Mythos 5: the same underlying model, but with some safeguards lifted.
- Access path: Mythos 5 initially shipped through Project Glasswing for a small group of cyberdefenders and infrastructure providers.
- Pricing: both models launched at $10 per million input tokens and $50 per million output tokens.
That release framing makes the reversal sharper. Three days earlier, Anthropic was describing Mythos 5 as a government-collaborative trusted-access program for cyber defenders; by June 12, the company said a US directive had forced both Mythos 5 and the public Fable 5 offline.
API access precedent
Discussion around Statement on US government directive to suspend access to Fable 5 and Mythos 5
Thread discussion highlights: - softwaredoug on API access restrictions: The US already has export controls for model weights. It appears this sets a precedent for even API usage being restricted. - w_t_payne on systemic dependence on cloud models: This really is a forcing function which makes us question if the risks associated with large "cloud-based" models are worth it, and if we need to find out if we can do more with smaller, local models. - jMyles on export-controls framing: If I make some statement, and post it on the internet, and someone downloads it in another country, I haven't exported anything.
Statement on US government directive to suspend access to Fable 5 and Mythos 5
The key engineering angle is that a government order has suspended access to a frontier model via API, which raises questions about compliance architecture, regional/customer gating, export-control scope, and operational risk for teams depending on hosted models. The thread also touches on cyber-safety evaluation thresholds, jailbreak detection, and the business risk of building on cloud-only model providers.
The most useful community reaction was not outrage about downtime. In the HN discussion, commenters treated the episode as a precedent question.
The thread pulled out three distinct concerns:
- API access restrictions: discussion highlights notes that commenters saw the order as extending export-control logic from model weights to hosted API usage.
- Cloud dependence: main HN thread overview highlights comments arguing that teams built on cloud-only frontier models inherit policy risk as well as technical risk.
- Anthropic's own safety politics: the thread overview also surfaced the irony that Anthropic has argued for stronger government powers to block unsafe model deployments, while now calling this action technically unfounded and procedurally unfair.
That last point is new because Anthropic's own statement did not just contest the evidence. It also said the government should have authority to block unsafe deployments, but only through a transparent statutory process grounded in technical facts.