US Government Forces Anthropic to Disable Advanced AI Models Worldwide

The US government has directed Anthropic to shut down access to its most powerful AI models, Fable 5 and Mythos 5, worldwide, marking the first known instance of the US government forcing a complete global shutdown of advanced AI capabilities. The action represents a significant escalation in government intervention over AI systems and raises questions about regulatory precedents for the entire industry.

Scope of the Directive

The export control directive bans all access to Fable 5 and Mythos 5 by foreign nationals, whether they're inside or outside the US, and affects even Anthropic's own foreign employees. To comply with this requirement, Anthropic has cut off access for all customers worldwide rather than attempting to implement country-specific restrictions. All other Anthropic models remain available, according to the company's statement.

The global nature of the shutdown distinguishes this action from traditional export controls, which typically restrict sales or transfers to specific jurisdictions. By forcing a worldwide blackout, the government has essentially prevented any use of these models regardless of where users are located or their citizenship status.

Government's Safety Rationale

According to Anthropic, the government believes it has found a method to bypass Fable 5's safety measures. The government claims this potential jailbreak presents a national security risk significant enough to warrant immediate model suspension.

The potential jailbreak—so far only described verbally by the government—boils down to asking the model to read a specific codebase and fix software bugs. This technique would theoretically allow users to extract capabilities from the model that its safety systems are designed to prevent.

Anthropic's Technical Pushback

Anthropic is complying with the order but has made clear its disagreement with the government's assessment. The company reviewed a demo of the technique and found it identifies only "a small number of previously known, minor vulnerabilities" that other publicly available models could also detect.

Anthropic says the capabilities shown are "widely available from other models," including OpenAI's GPT-5.5, and that security researchers already use these capabilities daily to protect systems. This argument undercuts the government's premise that Fable 5 presents unique risks unavailable elsewhere in the market.

The company emphasizes the extensive testing already conducted on these models. Before launch, the US government, the UK AI Safety Institute (UK AISI), private third-party organizations, and internal teams tested the model for thousands of hours combined, and the safety measures are "substantially more effective than those of any previously deployed model," Anthropic says. No tester has found a universal jailbreak, a method that could broadly bypass the model's safety measures and unlock a wide range of cyber capabilities.

The "Defense in Depth" Strategy

Anthropic has built its safety approach around the acknowledgment that perfect jailbreak resistance is not currently possible for any model provider. Every safeguard used across the industry is vulnerable to non-universal jailbreaks that can extract some information in specific cases, Anthropic says.

In response to this technical reality, the company pursued a strategy it calls "defense in depth": keep jailbreaks either narrowly scoped or expensive to pull off, combined with broad monitoring to quickly detect and shut down successful attacks. Part of this strategy includes 30-day data retention for customer data, which Anthropic says creates "real costs for us with customers" but enables jailbreak research and mitigation.

Industry Precedent Concerns

Anthropic is complying with the order but making its objections clear. "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", the company stated.

The company has articulated a critical concern about the broader implications of this action. If this standard were applied across the industry, it would effectively halt all new model deployments from every frontier model provider, the company says. This warning highlights how the government's action, if replicated for other advanced systems, could freeze AI model development at commercial scale.

Anthropic has previously articulated its preferred regulatory approach. In earlier public statements, Anthropic argued that the government should have the power to block unsafe deployments, but through a legal process that is "transparent, fair, clear, and grounded in technical facts." The current action doesn't meet those principles, the company says, hinting that this could become another chapter in the ongoing clash between Anthropic and the US government.

Regulatory Context

The US government recently issued a new executive order that lets AI developers submit their models for government safety review before release. Anthropic welcomed that approach, but the process apparently wasn't in place yet when the directive came down. This timing suggests a potential gap between the formalization of new oversight mechanisms and their practical implementation in regulatory decisions.

Persistent Industry Challenge

Jailbreaks and the related problem of prompt injections have been an unsolved security problem since the early days of large language models, and no LLM maker is immune. The vulnerability spans the entire industry and has affected systems from their inception. ChatGPT and Claude can still be attacked through prompt injection under certain conditions, even though their makers have added countermeasures.

About a year ago, Anthropic built a specialized defense against manipulation attempts and put it through a public jailbreaking challenge. After five days, over 300,000 messages, and roughly 3,700 collective work hours, the system was completely cracked, including a universal jailbreak. This history demonstrates that even targeted security efforts have consistently fallen short against determined adversaries, and that the problem of jailbreak resistance is fundamentally different from traditional cybersecurity challenges.

The shutdown of Fable 5 and Mythos 5 therefore reflects a government decision to restrict deployment of models with known but manageable vulnerabilities rather than waiting for perfect security solutions that the industry has not yet achieved. Whether this represents a new standard for AI regulation remains to be seen, particularly as Anthropic calls the move a "misunderstanding" and says it's working to restore access as quickly as possible.