US Imposes Export Controls on AI Models, Triggering Global Shutdown of Anthropic’s Top Systems

The US government has begun treating artificial intelligence models as strategic assets that require approval to cross borders, putting them in the same category as advanced semiconductors and military technology. The shift came as Washington moved to block overseas access to the latest models from a US AI company.

The Commerce Department issued export-control guidance on June 12 barring all foreign nationals from using Anthropic’s top-tier models, Claude Mythos 5 and Claude Fable 5. Anthropic immediately suspended both services for all customers worldwide to comply with the order. It is the first known case of a commercially deployed AI model being halted through direct federal intervention.

Anthropic said Commerce Secretary Howard Lutnick sent a letter that day to Chief Executive Officer Dario Amodei designating Mythos 5 and Fable 5 as subject to export controls in all locations outside the US and for all foreign nationals inside the country. The restrictions apply not only to overseas users but also to foreign nationals residing in the US. Anthropic’s foreign employees are included as well.

Any export, re-export or transfer of the models within the US now requires an individual license. Civil and criminal penalties apply to violations. Anthropic said it received the directive at 5:21 p.m. Eastern time on June 12. Because it had no way to distinguish foreign nationals from US citizens in real time, the company imposed a blanket suspension for all customers. Other models, including Opus, are unaffected.

Anthropic pushed back against the government even as it complied with the order. The company said it believes officials objected to a jailbreak technique that bypasses safeguards in Fable 5. It described the method as narrow and non-general-purpose, involving prompts that ask the model to read a specific codebase and fix flaws. In Anthropic’s view, the technique exposed only a few minor vulnerabilities that were already known.

The company also argued that similar capabilities could raise the same concerns for other public models not subject to export controls, including OpenAI’s GPT-5.5. Before launch, Anthropic said, the models underwent thousands of hours of red-team testing with the US government, the UK AI Security Institute and outside organizations. It added that no one has yet found a general-purpose jailbreak that broadly disables the safeguards.

Anthropic said the action stemmed from a misunderstanding and that it would work to restore service as quickly as possible. Recalling a commercial model distributed to millions of users over a single difficult jailbreak technique is unfair, the company said. If the same standard were applied across the industry, new model launches by every leading AI developer would effectively stop.

The move follows an escalating clash between Anthropic and Washington. Anthropic had already been placed on a Defense Department blacklist after the Pentagon determined the company was too risky even for internal government use. Now the Commerce Department has also put foreign use under a licensing regime. A US administration official told Axios that the government had tried and failed to delay the release of Anthropic’s latest model before sending the control letter. The block could last several weeks until national-security safeguards are strengthened, the official told Axios.

The official also said President Donald Trump does not want to damage the AI industry or stifle innovation. Earlier in June, the Trump administration issued an executive order requiring reviews of cutting-edge AI models before deployment. But that process was voluntary and did not involve licensing, making the new restrictions materially different.

The decision sent shock waves through the global AI industry. Dean Ball, an AI policy specialist who briefly served in the Trump administration and has recently criticized Anthropic’s policies, called the move “cartoonish” in a post on X. He wrote that it was contradictory to allow exports of advanced AI chips to China while barring citizens of allies such as the UK from using top-performance models.

Others argued Anthropic had helped bring about the outcome by previously describing Mythos as too dangerous for broad public release. Industry observers are focused on the wider implication: the logic of export controls has shifted from semiconductors and hardware to the models themselves. One analysis said the US government has started treating frontier AI as a strategic asset requiring licenses, much like advanced chips or military technology, setting a clear precedent for the industry.

The fallout in South Korea could also be significant. The guidance blocks all foreign nationals, shutting Korean companies, research institutes and government bodies out of Anthropic’s top-tier models. Mythos 5 had previously been supplied only on a limited basis to vetted institutions through Project Glasswing, which was operated with the US government. Organizations cited as participants included the Korea Internet & Security Agency, Samsung Electronics, SK Hynix and SK Telecom.

Even if those groups had secured access, that channel has now been cut off. South Korea is a key supplier to US AI infrastructure through the high-bandwidth memory chips made by SK Hynix and Samsung Electronics. Yet it now risks being barred as a user of the highest-performance models running on that hardware. Critics say the country can supply the chips but not use the AI brain.

The industry is now watching for the next step. The US administration has indicated the restrictions could be lifted within weeks. But if withdrawing a commercial model over a single difficult jailbreak case becomes a precedent, deployment strategies across the AI sector could be thrown into doubt. With OpenAI and Anthropic both pursuing public listings, government intervention in the name of safety could also affect valuations and global business plans.

Kim Ju-wan, Hankyung.com reporter kjwan@hankyung.com