The United States government issued an emergency export control directive on Friday, mandating that AI research firm Anthropic immediately suspend access to its two most advanced artificial intelligence models, Claude Fable 5 and Claude Mythos 5. Both models, which were reportedly released only days prior to the directive, are now inaccessible to all foreign nationals, including Anthropic’s own international employees. The government cited national security concerns as the primary justification for this unprecedented action, which has forced Anthropic to disable the models for its entire customer base to ensure compliance with the sweeping order.
The directive’s broad scope prohibits any foreign national from accessing these powerful AI systems, irrespective of their geographical location, whether within or outside the United States. This stringent measure has effectively compelled Anthropic to restrict access globally, a move that underscores the perceived gravity of the national security implications.
While the official directive did not provide granular specifics regarding the nature of the national security threat, sources indicate that the government has become aware of a method to bypass or "jailbreak" the publicly available Fable 5 model. The Mythos 5 model, which is characterized by fewer built-in safety guardrails and a pronounced capability in identifying cybersecurity exploits, was reportedly only available to a select group of partners. The revelation of a potential vulnerability in these cutting-edge AI systems has triggered an urgent governmental response.
Anthropic, in response to the directive, has expressed reservations about the severity of the identified vulnerabilities. The company stated that it reviewed a demonstration of the alleged bypass technique and concluded that the identified flaws appear relatively straightforward. Furthermore, Anthropic asserted that other publicly available AI models are capable of discovering similar vulnerabilities without requiring any specialized bypass methods.
According to Anthropic, the government has thus far presented only verbal evidence of what they describe as a potentially narrow, non-universal jailbreak. This technique reportedly involves instructing the AI model to analyze a specific codebase and identify any software flaws. Anthropic claims to have validated that the level of capability demonstrated in this bypass is already widely accessible through competing AI models, including those developed by OpenAI, such as GPT-5.5. This assertion suggests that the vulnerabilities, while present, may not be unique to Anthropic’s models and could be replicable across the industry.
Despite its disputations, Anthropic is adhering to the government’s directive. However, the company has voiced concerns that this action could establish a dangerous precedent for the entire AI industry. "If this standard was applied across the industry," Anthropic stated in a release, "we believe it would essentially halt all new model deployments for all frontier model providers." This sentiment highlights the potential chilling effect such measures could have on AI innovation and deployment, particularly for companies operating at the forefront of AI development.
Access to all other Anthropic models, which are not subject to the directive, remains unaffected. The company has indicated that it is actively working towards restoring access to Claude Fable 5 and Claude Mythos 5 as expeditiously as possible, contingent on resolving the concerns raised by the government.
Escalating Concerns and Public Statements
The situation took a more public turn with a series of statements made by David Sacks, co-chair of the President’s Council of Advisers on Science and Technology. On Saturday, Sacks posted on the social media platform X, formerly Twitter, providing an account of the events leading up to the government’s directive. He stated, "I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true."
Sacks elaborated that a "highly credible, trusted partner of both Anthropic and the U.S. government," while testing Fable, reportedly identified a jailbreak that circumvented the model’s safety guardrails. According to Sacks, the administration then requested that Anthropic CEO Dario Amodei either fix the vulnerability or temporarily deactivate the model. Sacks alleged that Amodei refused this request.
This alleged refusal, Sacks contends, directly contradicts Anthropic’s publicly stated commitment to AI safety and regulation, a stance that CEO Dario Amodei has recently reiterated in a blog post. Sacks posited, "Anthropic prioritized the continued offering of the consumer model over safety." He further claimed that the administration issued the export control directive "reluctantly" and expressed surprise at Anthropic’s perceived lack of cooperation with a "reasonable safety request." Sacks concluded that Anthropic’s reaction is "very much at odds with their branding and ethos as a safe AI research community."
Sacks also conveyed the administration’s hope that Anthropic will address the identified issue, allowing for the eventual reinstatement of Fable 5’s public release. He added that the administration values Anthropic’s technical capabilities and believes the current issue, though serious, is "easily resolved." The onus, he stated, is now on Anthropic to rectify the situation.
A History of Tensions Between Anthropic and the U.S. Government
This recent directive is not the first instance of friction between Anthropic and the U.S. government. Earlier this year, the AI firm reportedly refused to sign an expanded agreement that would have permitted extensive domestic surveillance of American citizens and the deployment of fully autonomous lethal weapons systems. This refusal led to a public rebuke from then-President Donald Trump, who criticized the company after negotiations stalled.
Subsequently, the Department of Defense designated Anthropic as a "supply chain risk," a label that the company has actively challenged in legal proceedings. Despite this adversarial period, reports emerged that the standoff between the two entities had begun to cool, with the government reportedly reconsidering the use of Claude Mythos and other Anthropic models. This recent export control directive, however, suggests that underlying concerns about AI safety and national security remain a significant point of contention.
Broader Implications for the AI Industry
The U.S. government’s emergency export control directive on Anthropic’s advanced AI models carries significant implications for the broader artificial intelligence landscape. This action highlights a growing governmental inclination to exert regulatory control over the development and deployment of powerful AI technologies, particularly those with potential national security ramifications.
The directive underscores the challenge of balancing rapid AI innovation with the imperative of national security. The government’s perceived need to intervene suggests that existing regulatory frameworks may be insufficient to address the evolving risks posed by frontier AI models. The government’s concern about a "jailbreak" method, even if deemed minor by Anthropic, indicates a low tolerance for potential vulnerabilities that could be exploited for malicious purposes, such as cyberattacks or the generation of harmful content.
Anthropic’s assertion that this action could "halt all new model deployments" if applied broadly is a critical point. It raises questions about the feasibility and practicality of such stringent export controls, especially for technologies that are rapidly advancing and have global reach. The ability of foreign nationals, including international employees of AI companies, to access these models is a complex issue with geopolitical dimensions. Restricting access based on nationality could hinder international collaboration in AI research and development, potentially impacting the U.S.’s competitive standing in the global AI race.
The situation also brings to the forefront the delicate interplay between private sector innovation and governmental oversight. While companies like Anthropic are pushing the boundaries of AI capabilities, governments are tasked with ensuring that these advancements do not pose undue risks to societal safety and national security. The differing perspectives on the severity of the identified vulnerabilities – with the government viewing them as a critical threat and Anthropic as a manageable issue – exemplify the ongoing debate about risk assessment and mitigation strategies in the AI domain.
Furthermore, the involvement of David Sacks, a prominent figure in technology and policy, as well as his public commentary, indicates a high level of government engagement with this issue. His statements suggest an administration that is actively monitoring AI development and is prepared to take decisive action when it perceives a threat. The reference to a "highly credible, trusted partner" also points to a collaborative approach between government agencies and industry stakeholders in identifying and addressing AI risks.
The previous tensions between Anthropic and the U.S. government over issues like surveillance and autonomous weapons suggest a pattern of disagreement on the ethical and security implications of advanced technologies. This latest incident reinforces the idea that the relationship between leading AI developers and government entities is likely to remain complex and dynamic, characterized by both cooperation and contention.
As the situation unfolds, the focus will be on Anthropic’s response to the government’s demands and the broader implications for AI regulation. The resolution of this directive could set a precedent for how the U.S. government handles future instances of alleged AI vulnerabilities and the extent to which it is willing to intervene in the development and dissemination of powerful AI models. The global AI community will be closely watching to see if this action leads to a more restrictive regulatory environment or if a balance can be struck between fostering innovation and ensuring responsible AI development. The outcome of this dispute will undoubtedly shape the future trajectory of AI governance and its impact on both national security and technological progress.
