U.S. Blocks Foreign Access to Anthropic's Advanced AI Models in Escalating National Security Dispute

The Trump administration has ordered the artificial intelligence company Anthropic to block foreign nationals from accessing its most advanced AI models, marking an unprecedented escalation in Washington's effort to treat cutting-edge AI systems as strictly controlled national security assets. U.S. Commerce Secretary Howard Lutnick issued a directive applying immediate export controls to Anthropic's newly developed Fable 5 and Mythos 5 models. The order explicitly prohibits the models from being accessed in any location outside of the United States, and further bans all foreign persons within the country from using the technology. In response to the sudden mandate, Anthropic took both models completely offline on Friday afternoon, cutting off all customer access to ensure compliance while it navigates the legal and technical requirements of the directive.[1][2][3]

The export controls represent the most significant step the U.S. government has taken to date to restrict global access to frontier AI technology. Anthropic had just released the Fable 5 model widely earlier in the week, while access to the more powerful Mythos 5 model had already been tightly restricted internally due to the company's own cybersecurity fears. According to statements from the company, the government's intervention stems from a belief that there is a method to "jailbreak" or bypass safeguards within Fable 5, potentially allowing the model to be used to identify critical software vulnerabilities. The Commerce Department has not publicly detailed the specific national security threats that prompted the emergency directive.[2][4]

Anthropic publicly disagreed with the administration's handling of the situation, characterizing the government's concerns as a "misunderstanding." In a statement released shortly after taking the models offline, the company argued that while the government should have the authority to block unsafe AI deployments, such actions must be part of a statutory process that is "transparent, fair, clear, and grounded in technical facts." Anthropic asserted that the sudden export control directive did not adhere to those principles, though the company expressed hope that it could restore access to the models as soon as possible once the dispute is resolved.[2][6]

The sudden imposition of export controls is not an isolated incident, but rather the latest flashpoint in a bitter, months-long standoff between the AI developer and the federal government. The relationship began to fracture in the fall of 2025 during negotiations over a $200 million contract with the Department of Defense. Anthropic's Claude models had previously been cleared for classified military use, but tensions escalated when the Pentagon demanded that the company allow the military to use the technology for "any lawful use." Anthropic drew firm ethical lines, refusing to allow its models to be deployed for fully autonomous lethal weapons or mass domestic surveillance.[5]

When Anthropic CEO Dario Amodei refused to yield on those safety guardrails, Defense Secretary Pete Hegseth issued an ultimatum, demanding compliance or severe consequences. In March 2026, the Pentagon formally designated Anthropic as a "supply chain risk" to national security. The designation, typically reserved for foreign adversaries with alleged ties to hostile governments, effectively blacklisted the American company from federal contracting. The Defense Department ordered all federal agencies to cease using Anthropic's technology and warned that any defense contractor continuing to do business with the company could face severe penalties.[5]

Anthropic responded to the blacklisting by filing a federal lawsuit in California, arguing that the designation was unlawful retaliation and a violation of its First and Fifth Amendment rights. The company accused the administration of weaponizing national security statutes to punish a domestic business for its ethical speech and safety policies. In response, the U.S. Department of Justice challenged the lawsuit on procedural grounds, arguing that the ban was not subject to court review because it did not constitute a "final agency action." The government denied any unlawful retaliation, though it acknowledged that agencies moved to cut off the company's products after it resisted Pentagon demands regarding military applications.[3]

The Pentagon's aggressive posture toward Anthropic sparked a rare and massive revolt across the broader technology industry. In an unprecedented show of solidarity, dozens of researchers from rival AI labs, including OpenAI and Google DeepMind, filed an amicus brief supporting Anthropic's legal challenge. The coalition warned that using national security labels to punish a company for implementing safety guardrails would create a severe chilling effect, deterring technical experts from speaking openly about AI risks for fear of federal retaliation. Furthermore, major cloud providers—including Amazon Web Services, Google Cloud, and Microsoft—reportedly signaled their defiance of the Pentagon's commercial blacklist, refusing to drop Anthropic's models from their enterprise platforms.[5]

The new export controls on Fable 5 and Mythos 5 arrive just ten days after President Trump signed an executive order establishing a framework for the federal government to vet the national security risks of advanced AI systems. That order proposed a voluntary system where AI developers would submit their most powerful models for up to a month of government review before public release. However, the mandatory and immediate nature of the Commerce Department's directive against Anthropic suggests that the administration is willing to use hard regulatory power to control the flow of AI technology when it deems voluntary cooperation insufficient.[2][6]

The escalating conflict places Anthropic in a precarious position as it attempts to navigate both the Pentagon's contracting blacklist and the Commerce Department's global export controls. The company, which had reportedly filed confidentially for a U.S. initial public offering earlier in the month, now faces immense uncertainty regarding its revenue streams and its ability to operate internationally. Beyond the immediate financial impact on a single company, the standoff serves as a landmark test of executive authority over the domestic technology sector, raising fundamental questions about whether private developers or the federal government will ultimately dictate the ethical boundaries and global distribution of artificial intelligence.[1][3]

The concept of a "jailbreak" in the context of large language models refers to specialized prompts or techniques designed to bypass the safety filters and ethical constraints programmed into the AI by its creators. If the government's intelligence suggests that foreign actors could reliably jailbreak Fable 5, the model could theoretically be manipulated to generate malicious code, identify zero-day exploits in critical infrastructure, or automate cyberattacks. By treating the model itself as a dual-use munition subject to export controls, the Commerce Department is attempting to physically and digitally quarantine the software, preventing it from crossing borders or falling into the hands of foreign nationals operating within the United States.[4]

Enforcing such a sweeping ban on foreign nationals presents a monumental technical and logistical challenge. Cloud-based AI models are typically accessed via application programming interfaces (APIs) that process millions of requests per minute from users around the globe. To comply with the directive, Anthropic would need to implement draconian identity verification systems, potentially requiring users to submit proof of U.S. citizenship before being granted access to the models. Because building and deploying such an infrastructure overnight is virtually impossible, the company was forced to take the nuclear option of pulling the models offline entirely, disrupting service for domestic and international users alike.[1][2]

As the legal battles consolidate in federal courts in California and Washington, D.C., the broader artificial intelligence industry is watching closely. The outcome of Anthropic's clash with the Trump administration will likely establish the legal precedent for how the U.S. government regulates frontier AI models for the next decade. If the courts uphold the administration's use of supply chain risk designations and emergency export controls, AI developers may be forced to choose between adopting the Pentagon's unrestricted terms of use or facing exclusion from the global market. For now, the most advanced models produced by one of America's leading AI labs remain locked away, caught in a high-stakes tug-of-war between Silicon Valley's safety culture and Washington's national security imperatives.[3][5]