Summary of the Article
- The US government has ordered Anthropic to suspend foreign national access to its two most advanced AI models, Fable 5 and Mythos 5, due to national security concerns related to cybersecurity vulnerabilities.
- Anthropic has suspended access for all users worldwide — not just foreign nationals — as it had no reliable way to verify nationality at scale without shutting everything down.
- The Mythos model reportedly raised concerns among US officials due to its ability to exploit cybersecurity weaknesses at an unprecedented rate, prompting one of the most comprehensive AI export control actions ever taken.
- Anthropic has publicly disagreed with the order, arguing that it is an unfair standard that, if applied across the industry, would force similar shutdowns across every major AI provider.
- Continue reading to understand what this means for the future of AI regulation, Anthropic’s upcoming IPO, and whether this type of export control could become the new normal.
A single government order has taken two of the world’s most advanced AI models offline for every user worldwide.
The US government ordered Anthropic, the AI safety and research company responsible for the Claude series of models, to restrict access to Fable 5 and Mythos 5 for all non-US citizens. The issue? It’s not technically possible to verify nationality on a large scale, in real time, across a worldwide user base. So Anthropic did the only thing it could: it completely shut down both models. For everyone.
This is one of the broadest restrictions on AI access that the US government has ever imposed on a domestic AI company. It was issued without a detailed explanation of the specific national security concern, and it was announced just days after Anthropic had publicly called for more government regulation of AI, a position that now seems bitterly ironic.
The US Government Has Closed Access to Anthropic’s Most Advanced AI Models
Anthropic released a public statement confirming the closure, stating that the US government has issued an export control order that requires the company to suspend access to Fable 5 and Mythos 5 for all non-US citizens. Anthropic stated that it was not provided with specific details about the national security concern that led to the order.
Anthropic’s Fable 5 and Mythos 5 Are No Longer Accessible to Users
Anthropic launched Fable 5 just a few days before the order. It was a general-use version of their most powerful model tier, Mythos. According to Anthropic, their Mythos-class models represented the highest level of AI capability. The order did not affect access to any other Anthropic models — only these two. For more information on the impact of this withdrawal, you can read about the US government’s order.
Both consumer products and enterprise API access are affected by the closure. Developers, businesses, and individual users who had incorporated Fable 5 into their workflows suddenly lost access, with no transition period provided.
The term “abruptly” in that statement is quite significant. There was no phased launch, no warning period for developers to shift, and no grandfathering of existing business agreements. The order was issued, and the models were shut down, as detailed in the impact of the US government’s decision.
Anthropic’s reasoning behind the global access restriction, not just for foreign nationals
Although the order specifically targeted foreign nationals, the global user base of Anthropic makes this distinction almost impossible to enforce in reality. There is no existing plug-and-play solution for real-time nationality verification at the scale of an API. To avoid the risk of non-compliance, or even worse, selectively enforcing a rule that cannot be enforced, Anthropic decided to take both models completely offline for all users.
What Does the Export Control Directive Really Mean?
Usually, the Department of Commerce in the US issues export control directives. These directives are intended to stop sensitive technology from falling into the hands of foreign adversaries. The fact that this framework is now being applied to an AI model that is publicly accessible, rather than hardware or classified software, marks a major shift in the way these tools are being utilized.
The mandate explicitly pointed out Fable 5 and Mythos 5, hinting that the government has established guidelines to recognize particular AI models as export-sensitive based on their abilities, not on their deployment context. This is a significant difference. Earlier export controls in tech have usually focused on physical components such as semiconductors, not on software models that can be accessed via API.
Dean Ball, who used to work in the White House and helped create the AI Action Plan that was released in the summer of 2025, has spoken out about the order. He said that it essentially stops anyone who isn’t American from using the newest models from Anthropic. This includes people from other countries who are currently in the United States.
The Commerce Department’s Involvement in the Order
The Commerce Department has gradually become the main regulatory entity for national security issues related to AI. This order is part of a larger trend of the department extending its authority to AI model capabilities, going beyond the usual export controls focused on hardware.
The interesting part about this case is the lack of detail. Usually, export control orders are given for a specific vulnerability, threat actor, or classified finding. Anthropic has said outright that it was not given any specifics about the national security concern. This means that the company followed an order that it could not fully evaluate or challenge on technical grounds.
Policy observers have criticized the lack of transparency, arguing that due process requires at least some explanation of the reasoning behind such broad restrictions.
| Model | Status | Reason | Other Models Affected? |
|---|---|---|---|
| Fable 5 | Completely disabled | Export control directive | No |
| Mythos 5 | Completely disabled | Export control directive | No |
| All other Claude models | Not affected | Not named in directive | N/A |
What “Foreign Nationals” Means for Everyday Users
According to export control law, “foreign national” generally refers to anyone who is not a US citizen or lawful permanent resident — regardless of where they currently live or work. This means a French software engineer working at a US company on a work visa could technically fall under the restriction. It also means verifying compliance would require identity and citizenship checks that most AI platforms simply are not built to perform.
How Amazon Web Services Got Involved in the Shutdown
Anthropic has a significant cloud partnership with Amazon Web Services, which allows enterprise customers to access its models. Because AWS is involved, the shutdown had a ripple effect on enterprise deployments that were running Fable 5 or Mythos 5 through Amazon’s infrastructure. This added another level of disruption, affecting more than just Anthropic’s direct customers. For more on how Anthropic is impacting business automation, check out this comparison of business automation solutions.
The Catalyst That Led to the Government’s Decision
The government’s worry seems to stem from the Mythos model tier’s cybersecurity capabilities. Before the official order, Anthropic had been privately informing US government officials about Mythos, explaining that it has an unparalleled potential for carrying out cyberattacks — specifically, its ability to find and take advantage of software vulnerabilities at a pace and size that current defenses cannot consistently keep up with.
Both The Washington Post and The Wall Street Journal have reported that Mythos has prompted US officials to adopt a more immediate regulatory stance. The model’s capacity to identify and take advantage of small coding errors — a process known as vulnerability discovery among security researchers — has been highlighted as especially hazardous in the realm of critical infrastructure protection.
Understanding the Concept of a Jailbreak
In the AI world, a jailbreak refers to a method that circumvents the safety measures implemented in a model, allowing it to generate outputs that it was programmed to reject. Anthropic, for its part, forms specialized teams to rigorously test its models against these methods prior to launching them. However, the company has admitted that these protective measures are not foolproof. No security measure is invincible, and a determined foe with ample time can usually find a way around it.
Anthropic’s Take on the Threat
Anthropic has a different perspective on the situation, arguing that the threat posed by Mythos has been blown out of proportion. The company reminded the public that it had worked closely with the US government to establish safety protocols prior to the launch of Fable 5. The joint reviews didn’t reveal any findings that warranted such drastic restrictions. According to Anthropic, while the model is advanced, it doesn’t offer anything radically different from what other cutting-edge AI systems are already capable of.
What Other AI Models Can Do
Anthropic’s argument gets really interesting here. The company pointed out that other AI models from other major companies can do the same thing. They can find small bugs in code. This is the same ability that seems to have worried the government. If the government is going to restrict exports based on this ability, then it should apply to all AI models, not just Anthropic’s. Anthropic pointed this out in their public statement.
Anthropic Challenges Government’s Decision
Anthropic did not resist the order. There’s no debate about that. But the company made a conscious decision to comply in a very public and outspoken manner — putting out a statement that questions the government’s logic rather than silently taking the models offline without a word. This mix of obedience and critique is not common, and it suggests that Anthropic is viewing this as a policy fight, not simply a business interruption.
Anthropic said in no uncertain terms that the government’s move was not in keeping with the principles of fair and fact-based regulation. This was a sharp rebuke, especially since Anthropic had only a few days before called for that very type of regulation from the government. The juxtaposition was stark: an AI firm calling for regulation, then immediately being subjected to what it saw as unfair regulation.
Anthropic’s Compliance Despite Disagreement Explained
The reason is simple: severe legal consequences are associated with export control violations, and no company, no matter how much it disagrees with a directive, can afford to operate outside of federal compliance requirements. There was no practical way for Anthropic to contest the order before the deadline, and partial compliance was not technically possible due to the nationality verification issue. The only defensible option was a complete shutdown.
Anthropic’s Prediction for the Future of AI if This Standard Is Applied Across the Board
Anthropic is essentially saying that if the government applies this standard consistently, it would have to shut down all major AI providers operating at the frontier. OpenAI’s GPT-4o, Google’s Gemini Ultra, and Meta’s Llama models have all shown that they can analyze code and identify vulnerabilities. If the ability to handle cybersecurity is enough to trigger an export control directive, the US AI industry will have to choose between building capable models and keeping them commercially available.
Anthropic’s argument has wide-ranging implications. They are saying that the government has created a standard that it can’t selectively apply. If it applies it broadly, it would hamstring the US AI industry’s ability to compete on a global scale. This is in direct conflict with the administration’s stated goals of dominating in AI.
US Focuses on Protecting Critical Infrastructure
The US government is primarily concerned about AI-assisted attacks on critical infrastructure. This includes power grids, water systems, financial networks, and local government IT systems. These are all high-value targets that could suffer catastrophic damage if automated vulnerability discovery tools are used against them. The Mythos model can reportedly exploit software weaknesses quickly, which is especially concerning because critical infrastructure often relies on outdated software with known vulnerabilities that haven’t been patched.
The Vulnerability of Local Governments
State and local government systems are generally seen as the weakest link in the chain of critical US infrastructure. The software used by many local governments is out of date, and they often have small budgets for IT security, little to no patch management, and no dedicated cybersecurity personnel. An AI model that can quickly find weaknesses in old code poses a much bigger threat to these systems than it does to enterprise environments with more resources. The government is less worried about sophisticated enemies, who already have advanced hacking tools, and more worried about making it easier for less skilled threat actors to cause serious harm.
JD Vance’s Response to AI Company Executives
Vice President JD Vance was in contact with AI company executives during the time of this directive, and his public stance reflected the administration’s dual tension around AI: aggressive promotion of US AI dominance on one hand, and escalating concern about the national security implications of the most capable models on the other. The administration has consistently pushed a narrative that the US must lead in AI development — making a directive that hobbles a leading US AI company’s flagship product a difficult message to square publicly.
Politics plays a significant role in this situation. Anthropic’s ties with the government were already strained in 2025 when the company declined the US military’s request to use its models for domestic surveillance and fully autonomous weapons systems. The government’s response was to put Anthropic on a supply chain blacklist. The export control directive on Fable 5 and Mythos 5 comes in the midst of this already strained relationship, making it difficult to distinguish between legitimate national security concerns and institutional friction.
This Order Represents a Significant Change in US Export Control Policy
Historically, US export controls have focused on tangible items — semiconductors, weapons systems, dual-use hardware. The expansion of this framework to software models that can be accessed via API signifies a fundamental change in the government’s approach to technology control. While a chip can be traced, a model weight file can be duplicated, transmitted, and implemented anywhere in the world in a matter of seconds. The methods used to enforce hardware controls simply do not apply to software in the same way.
Why export controls don’t work for AI models: AI model weights aren’t like physical parts. They’re digital files that can be copied and shared over and over again without losing any quality. Once someone has access to a model’s weights or they’re leaked, there’s no way to get them back. Export controls that only apply to API access only stop one way of distributing the weights and leave all the others wide open.
This difference is crucial for policy formulation. Prohibiting foreign nationals from accessing Fable 5’s API does not stop a resolute enemy from obtaining model weights by other methods — whether through spying, insider access, or collaborating with open-weight alternatives that offer similar capabilities. The directive tackles a superficial access point, leaving the underlying vulnerability unaddressed.
What this order does is set a precedent: the US government is now willing to use its export control authority to restrict access to AI models based on their capability profile, not just the context in which they are deployed. This precedent will influence how every frontier AI company designs, releases, and controls access to its most capable systems in the future.
Anthropic isn’t the only one feeling the effects. Every cutting-edge AI lab must now consider the chance that a model with enough capability could prompt a similar order — without any prior notice, detailed reasoning, or transition period. This consideration will impact release plans, capability benchmarking disclosures, and decisions about what capabilities to include in general-release models versus restricted tiers.
Trump’s Administration Proposes FDA-Like Regulations for AI
The administration is considering an FDA-like regulatory framework for AI. Under this framework, the most powerful models would need to be reviewed and approved before they could be made public. The Fable 5 and Mythos 5 situation is a prime example of why this framework is under consideration. The administration seems to want a way to review cutting-edge AI models before they are released to the public, especially when these models have obvious dual-use potential in cybersecurity or weapons development, just as the FDA regulates access to pharmaceuticals based on demonstrated risk profiles.
The main issue is that the speed of AI development is much faster than the pace the FDA model was built for. It takes years for a drug to be approved. On the other hand, leading AI models are being launched in a matter of months, with unpredictable capability leaps, even for the companies that create them. No one has yet figured out how to design a regulatory review process that is both thorough enough to be meaningful and quick enough to not just give up the frontier to less regulated competitors, as seen in the Anthropic partnership controversy.
Anthropic’s Plans to Go Public Meet Political Resistance
Anthropic’s plans to go public are not sitting well with the current government friction. A company that is preparing to go public needs to show regulatory stability, predictable revenue, and a clear path to commercial scale. Having two of their main models taken offline by a government order — without a detailed explanation, without transition time, and in the middle of an already rocky relationship with the government — introduces the kind of regulatory risk that institutional investors take seriously. The supply chain blacklist that the government put Anthropic on earlier in 2025 adds to this. These are not the conditions a company wants when it is about to debut on the public market.
There’s also a question of value. The value of Anthropic is heavily tied to its most capable models. Fable 5 and Mythos 5 represent the commercial peak of what the company can offer to enterprise customers. Turning off that peak, even if only temporarily, has direct implications for revenue — and the uncertainty around when or if those models will return to full availability makes it hard to confidently model future earnings. For investors evaluating Anthropic’s IPO, the question is no longer just about market share in AI. It’s about whether the company’s most valuable assets can be turned off by regulatory decree at any time.
The Future of Anthropic and the AI Sector
Anthropic is left with a few clear choices. The firm could continue to communicate with the government through official channels, trying to make a technical argument that the capabilities of Mythos 5 do not reach the level that would necessitate export controls under any consistent standard. It could strive to establish some sort of nationality verification system that would allow it to restore access for verified US users while remaining compliant. Alternatively, it could wait for the political and policy environment to change — a passive strategy that comes with a significant commercial price.
Everyone in the wider AI industry is keeping a close eye on this situation. The precedent being set here — that a frontier AI model can be restricted by an export control directive based solely on its capability profile, with minimal transparency — is something that every major AI lab will now have to take into consideration in their planning. We can expect to see changes in how companies disclose capability benchmarks, how they tier access to their most powerful models, and how aggressively they brief government officials ahead of major releases. Anthropic’s private government briefings on Mythos, which were intended to build trust and demonstrate responsibility, may have inadvertently sped up the very regulatory action the company was trying to avoid.
Common Queries
These are responses to the most frequently asked questions about the US government’s order and its implications for Anthropic, its users, and the broader AI sector.
Understanding Fable 5 and Mythos 5
Anthropic’s AI models, Fable 5 and Mythos 5, are the most advanced in the company’s lineup. Anthropic defines the Mythos tier as a new level of AI performance, significantly surpassing its previous models. Mythos 5 is the core research-grade model, and Fable 5 is a safer, more broadly applicable version intended for wider commercial use.
Both models are constructed using Anthropic’s Constitutional AI framework and come equipped with several layers of safety precautions. Notable features of the Mythos model tier include:
- Advanced code analysis: Ability to identify and explain vulnerabilities in complex codebases at a speed that significantly exceeds human review
- Cybersecurity dual-use capability: Can perform both defensive security analysis and, when guardrails are bypassed, offensive vulnerability identification
- Mythos-class reasoning: Anthropic’s designation for a new ceiling of multi-step reasoning and problem-solving performance
- Built-in use restrictions: Guardrails specifically blocking deployment in high-risk domains including autonomous weapons systems and domestic surveillance
- Enterprise API access: Available through direct Anthropic API and AWS infrastructure for enterprise customers
Fable 5 was positioned as the version safe enough for general commercial release, with Mythos 5 reserved for a more controlled group of key partners through a program Anthropic called Glasswing — focused specifically on securing critical software systems.
Why Did Anthropic Disable Access for All Users Instead of Just Foreign Nationals?
Anthropic did not have a technically reliable way to verify the nationality of its users on a large scale in real time. Export control compliance requires a high level of certainty — partial or best-effort verification is not legally enough when the penalty for non-compliance is severe. Faced with an order it could not selectively enforce without a significant risk of violation, Anthropic chose to completely shut down as the only compliant path available.
What is an AI Jailbreak and Why is it Important?
An AI jailbreak is a method, typically a meticulously crafted input or series of prompts, used to circumvent the safety measures integrated into a model and induce it to generate outputs it was programmed to reject. Jailbreaks are significant because even the most robust safety systems have vulnerabilities, and these vulnerabilities can be manipulated to retrieve harmful content, such as comprehensive guides for cyberattacks, building weapons, or other damaging uses. Understanding these vulnerabilities is crucial, as highlighted in projects like the AI Era Software Security Project Glasswing Guide, which aim to enhance security measures.
Anthropic admits that its model defenses, while thoroughly tested, are still vulnerable. The company employs specialized red teams whose only task is to find and fix these vulnerabilities before and after they are released. The problem with Mythos is that its advanced capabilities mean that a successful jailbreak could produce much more dangerous outputs than a less capable model would allow — increasing the risk of every safety failure.
Could This Order Be Applied to Other AI Companies Like OpenAI?
Yes — and Anthropic’s own public statement implies exactly that. If the standard for triggering an export control directive is the ability to identify software vulnerabilities at speed, then OpenAI’s GPT-4o, Google’s Gemini models, and Meta’s Llama architecture all demonstrate comparable capabilities. Anthropic explicitly noted that rival models show similar vulnerability-surfacing ability. Whether the government applies this standard consistently, or whether Anthropic is being targeted selectively due to its already-damaged relationship with the administration, is a question the industry is actively asking.
When Will Access to Fable 5 and Mythos 5 Be Restored?
Anthropic has not yet given a timeframe for when access will be restored. In order to fully restore access, one of three things will likely need to happen: the government will need to rescind or modify its directive, Anthropic will need to implement a nationality verification system that meets compliance requirements, or there will need to be a legal or policy challenge that changes the framework of the order.
Creating a verification system is technically challenging and would need a substantial investment in infrastructure. There is no ready-made solution to the problem of designing a system that can accurately differentiate between US individuals and foreign nationals, whether it’s for consumer accounts, enterprise API keys, or AWS-mediated access.
There are several hurdles to overcome in the policy challenge approach. Anthropic would have to prove in a court of law that the directive was issued without adequate factual basis or due process, which is a tall order when national security is at stake. The company has indicated that it believes the order was unjust and did not adhere to the principles of fact-based decision-making, but expressing disagreement and successfully challenging a national security directive in court are two entirely different things.
It’s clear that the sooner Anthropic can resolve this issue, the better for them — not just for their commercial standing but for their IPO prospects as well. For each week that Fable 5 and Mythos 5 are offline, enterprise customers are considering other options, developers are creating integrations with competing models, and institutional investors are reevaluating the risk profile of a company that can have its flagship products shut down by a regulatory order without any prior notice. Anthropic is one of the most closely monitored AI safety companies globally, and for anyone keeping an eye on what responsible AI development looks like when under real-world government pressure, this situation is a crucial case study.
Please provide the content that you want to be rewritten.
