Anthropic disabled Fable 5 and Mythos 5 after a U.S. export control directive restricted access to the models for foreign nationals. The shutdown raises a central frontier AI governance question: who decides when an advanced model is too risky to remain available after launch?

The directive affects Anthropic customers, foreign national employees at Anthropic, enterprise AI buyers, AI labs, cloud providers, and policymakers watching how cybersecurity concerns may justify model restrictions after deployment. Anthropic says the government cited national security authorities but did not provide specific details, and that its understanding is that the concern is tied to a narrow, non-universal jailbreak involving cybersecurity tasks.

Reuters reported that Amazon CEO Andy Jassy was among technology leaders who raised security concerns with senior Trump administration officials about Anthropic’s most advanced models. Amazon did not confirm whether it discussed Anthropic’s models specifically, telling Reuters that governments sometimes seek its counsel on potential security risks.

In short, the Fable 5 and Mythos 5 shutdown is a test of frontier AI oversight after deployment. The larger issue is what evidence, standards, and oversight process should be required before a government restricts access to a deployed frontier model.

Frontier AI governance refers to the rules, technical evidence, oversight processes, and decision authority used to determine when advanced AI systems can be released, restricted, monitored, or withdrawn.

The Anthropic Fable 5 and Mythos 5 shutdown is a frontier AI oversight case involving government intervention, cybersecurity safeguards, and unresolved questions about post-deployment model access.

Anthropic says the U.S. government issued an export control directive requiring the company to suspend access to Fable 5 and Mythos 5 for foreign nationals, whether inside or outside the United States. The restriction also applied to foreign national employees at Anthropic.

The company says it received the directive Friday at 5:21 p.m. ET and that the order’s practical effect was to force it to disable both models for all customers in order to ensure compliance. Access to other Anthropic models was not affected.

The directive came only days after Anthropic rolled out Fable 5 and Mythos 5. Reuters reported that Mythos had previously been held back from wide release because of hacking-capability concerns, while Fable was released publicly with cybersecurity safeguards.

Anthropic says the directive did not provide specific details about the government’s national security concern. The company says its understanding is that the government had become aware of a method for bypassing, or “jailbreaking,” Fable 5. Jailbreaking refers to techniques that attempt to get an AI model to bypass its safeguards and produce outputs the system is designed to restrict.

The publicly reported cybersecurity concerns also extended beyond Anthropic’s own account of the directive. Reuters reported that Amazon CEO Andy Jassy was among technology leaders who raised security concerns with senior Trump administration officials about Anthropic’s most advanced models. Amazon did not confirm whether it discussed Anthropic’s models specifically, telling Reuters that governments sometimes seek its counsel on potential security risks.

Ars Technica reported that the administration was concerned by reports of a jailbreak that could bypass broad classifier-based safeguards designed to block certain prompts involving cybersecurity, chemistry, and biology. Ars Technica also reported, citing Axios, that the administration requested a pause in the release of the models to allow the national security apparatus to harden against the threat.

Anthropic describes the reported issue more narrowly. The company says it reviewed a demonstration of the specific technique being used to identify a small number of previously known, minor vulnerabilities. Anthropic says those vulnerabilities appeared relatively simple and that other publicly available models can find those same vulnerabilities without requiring the same bypass.

Anthropic also says the government has given it only verbal evidence of a potential narrow, non-universal jailbreak. According to Anthropic, the reported technique “essentially consists of asking the model to read a specific codebase and fix any software flaws.”

In practical terms, Anthropic is describing a task-specific bypass rather than a broad failure of the model’s safeguards. The company’s argument is that this kind of technique produced limited cybersecurity findings and did not show a general method for unlocking restricted capabilities across the model.

The key point: the governance question depends on what kind of jailbreak should justify removing a deployed frontier model from the market. Anthropic distinguishes between a universal jailbreak, which could broadly bypass safeguards and unlock many restricted capabilities, and a narrow non-universal jailbreak, which may produce limited outputs in specific circumstances.

Anthropic says no testers have found a universal jailbreak for Fable 5, which is central to its argument that the reported technique does not justify recalling the model.

Anthropic is complying with the directive while disputing the standard it believes the government is applying. The company wrote, “We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users.” Anthropic also wrote, “However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”

Anthropic argues that Fable 5’s safeguards were extensively tested before release. The company says it worked with the U.S. government, the U.K. AI Safety Institute, multiple private third-party organizations, and internal teams to red-team Fable’s safeguards for thousands of hours.

The company says those tests showed that Fable’s safeguards were substantially more effective than those of any previously deployed model. It also says many users complained that the safeguards were overly broad.

Anthropic’s defense rests on a defense-in-depth strategy. The company says perfect jailbreak resistance is not currently possible for any model provider because every safeguard used in the industry is vulnerable to narrow, non-universal jailbreaks in some circumstances.

Anthropic says its goal was to make jailbreaks either narrow or expensive to produce, then pair those safeguards with monitoring that could detect and shut down successful attacks. Essentially, that means the company is arguing for layered risk reduction rather than a guarantee that no jailbreak can ever occur.

That strategy is also why Anthropic required 30-day retention of customer data for Fable. Anthropic describes that retention requirement as a policy change that comes with real customer costs, but allows the company to research and mitigate jailbreaks.

Anthropic also says it has not received a disclosure of a concerning non-universal jailbreak that led to a harmful result. The company says it reviewed a report it believes formed the basis of the government’s directive and found that the level of capability displayed there is widely available from other models, including OpenAI’s GPT-5.5, and is used every day by defenders working to keep systems safe.

Anthropic says it stands by its defense-in-depth approach and believes Fable’s risks are comparable to risks from existing models already deployed across the industry. The government has not publicly provided a detailed technical explanation of why Fable 5 and Mythos 5 required this level of restriction.

Reuters reported that Amazon CEO Andy Jassy was among technology leaders who raised concerns to senior Trump administration officials about security risks in Anthropic’s most advanced AI models. Reuters cited a person familiar with the matter.

Amazon did not confirm whether it spoke to government officials about Anthropic’s models. An Amazon spokesperson told Reuters, “As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks. When they occur, we don’t share the details of these discussions.”

That leaves an important governance question unanswered. Reuters’ reporting does not establish whether Amazon proactively raised concerns, whether government officials sought Amazon’s input, or what role Amazon’s feedback played in the final decision.

The transparency issue is how outside concerns are evaluated when a major technology company, a frontier AI lab, and government officials are involved in a decision that removes access to deployed AI models. Without a clear process, companies and customers have limited visibility into whether intervention is based on technical evidence, national security judgment, outside input, or some combination of those factors.

For future frontier AI oversight, that process matters. If cloud providers, security partners, or technology companies with commercial interests in AI infrastructure and model access are consulted on model risk, companies and customers should expect clear rules for how that input is evaluated, documented, and separated from competitive or commercial interests.

Reuters reported that The Information, citing a U.S. official, later reported that the administration was unlikely to force other AI firms to follow restrictions similar to those placed on Anthropic. Reuters said it could not immediately verify the administration’s plans for regulating other firms.

That reporting raises another unanswered governance question: why were restrictions imposed on Anthropic’s models specifically?

Reuters also reported that White House adviser David Sacks wrote in a social media post that officials issued the export control “reluctantly” after Anthropic CEO Dario Amodei “refused” to “fix the jail break or de-deploy the model.” Sacks also wrote that the administration hoped Anthropic would remediate the safety issue, the export control would be lifted, and Fable would return to general release.

That statement explains the administration’s stated reason for using the export control after Anthropic did not take the requested action. It still does not publicly explain the technical evidence behind the directive, the standard used to judge the reported jailbreak, whether competing frontier models were evaluated under the same criteria, whether those criteria are being applied across the industry, or how consistency is being determined.

Anthropic says it supports government authority to block unsafe deployments, but argues that such decisions should follow a clearer process. The company wrote, “As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.”

Anthropic argues that the model capability displayed in the report it reviewed is widely available from other models and used by defenders who keep systems safe. The company also wrote that if the same standard were applied across the industry, it believes the result would “essentially halt all new model deployments for all frontier model providers.”

The scope of the restriction also raised questions outside Anthropic. Reuters reported that some experts who favor export controls on advanced AI models found the action puzzling because it affected allied nations as well as adversaries. Jimmy Goodrich, a senior fellow at the University of California’s Institute for Global Conflict and Cooperation, told Reuters, “This was not well thought-out,” adding, "It even bans Canadians and Brits employed at Anthropic from doing research and development.”

The Fable 5 directive leaves the article’s central governance question unresolved. If governments can restrict deployed frontier models after launch, AI labs and enterprise customers will need a clearer process for deciding what evidence justifies restricting access to a deployed model, who gets consulted, and whether the same rules apply across AI providers.

Q: What happened to Anthropic’s Fable 5 and Mythos 5?
A: Anthropic disabled Fable 5 and Mythos 5 for all customers after the U.S. government issued an export control directive restricting foreign-national access to the models. Anthropic says it shut off the models broadly to ensure compliance with the directive.

Q: Who lost access to Fable 5 and Mythos 5 after the U.S. directive?
A: Anthropic says the directive required it to suspend access to Fable 5 and Mythos 5 for foreign nationals, whether inside or outside the United States. Anthropic says that restriction also applied to foreign national employees at Anthropic, so the company removed access for all customers as the immediate compliance step.

Q: Why was the U.S. government concerned about Fable 5?
A: Anthropic says the government cited national security authorities and appears to believe it had become aware of a method for jailbreaking Fable 5. Anthropic says the evidence it reviewed involved a narrow technique used to identify minor, previously known software vulnerabilities.

Q: What is a jailbreak, and why does the type of jailbreak matter?
A: A jailbreak is a technique that attempts to bypass an AI model’s safeguards. Anthropic says the reported concern involved a narrow non-universal jailbreak, not a universal jailbreak that could broadly unlock restricted capabilities. That distinction matters because the dispute centers on what kind of safeguard failure should justify removing a deployed frontier model.

Q: What is still unclear about Amazon’s role?
A: Reuters reported that Amazon CEO Andy Jassy was among technology leaders who raised concerns with senior U.S. officials about security risks in Anthropic’s advanced models. Amazon did not confirm whether it discussed Anthropic’s models specifically, and it remains unclear whether Amazon proactively raised concerns, whether officials sought Amazon’s input, or how any feedback affected the final decision.

Q: Why does the Anthropic shutdown matter for other frontier AI companies?
A: The shutdown raises frontier AI oversight questions because Reuters reported that similar restrictions may not be expected for other AI firms. Officials have not publicly explained whether competing frontier models were evaluated under the same standards, whether the same criteria are being applied across the industry, or how consistency is being determined.

The Anthropic case turns post-launch model access into a governance decision that extends beyond a company’s product release process. A model can pass internal testing, reach customers, and still face government restrictions if officials conclude the risk is significant. That does not mean government intervention was unwarranted. National security officials may have access to nonpublic evidence that companies, customers, and the public cannot fully evaluate.

The key issue is the standard for intervention. Anthropic does not deny that safeguards can be bypassed in limited cases. It argues that the reported technique was narrow, produced minor findings, and involved capabilities the company says are already available in other public models.

AI labs and enterprise customers should pay close attention because post-deployment oversight can affect model availability after adoption. Cloud providers and security partners should also watch the process because the case raises questions about how outside technical input, security expertise, and government review may interact when officials evaluate frontier model risk.

The timing matters because frontier AI models are becoming more relevant to coding, cybersecurity, scientific work, and autonomous workflows. As those systems become more capable, governments may be less willing to leave deployment decisions entirely to AI labs.

The practical decision is how much evidence, transparency, and consistency should be required before a government restricts access to a frontier model. AI companies will likely want clearer standards for what justifies intervention, who is consulted, and whether the same evaluation process applies consistently across providers.

In short, the Fable 5 and Mythos 5 shutdown turns model safety from a company release decision into a public governance question. Anthropic says the action was not transparent, fair, clear, or grounded in sufficient technical facts. The government has not publicly explained the full evidence behind the directive.

When governments decide a frontier model must come down, trust will depend on more than the model’s safeguards. It will depend on the credibility of the process.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.