The U.S. Department of Commerce has Lifted export controls that forced Anthropic to impose The company on Tuesday said it will shut down two of its most advanced artificial intelligence (AI) models – Fable 5 and Myth 5 – to all users globally, including those in India.
The reversal caps an incident that began on June 12, when the Trump administration ordered Anthropic to refuse to provide models to all foreigners, citing national security concerns. The order comes as Washington has been increasing oversight of advanced artificial intelligence models over concerns they could be misused for military intelligence by China, Russia or other interested countries.
Because the company couldn’t reliably separate foreigners from other users in real time, it turned off access to the model to everyone, including Americans.
Fable 5 was released in early June as a locked public version of Fable 5 that Anthropic withheld from the wider public due to the model’s unusually strong ability to spot vulnerabilities that hackers could exploit. Mythos 5 is only available to a small group of companies and governments as part of what Anthropic calls Project Glasswing, allowing these select groups and entities to discover other potentially dangerous uses or vulnerabilities in their own infrastructure.
Artificial first recovery Myth 5 has limited access On June 26, the U.S. Department of Commerce completely lifted restrictions on both models and subsequently made such services available to a small group of “trustworthy” U.S. organizations on June 26.
However, the restrictions have had ripple effects throughout the AI industry and raised questions about whether export controls can be used to regulate access to AI models.
What happened?
According to Reuters, Anthropic said the U.S. government’s export control order was issued after Amazon researchers found a way to bypass security measures in “Fable 5”, which allowed powerful artificial intelligence models to identify software vulnerabilities or access other illegal information.
The company said it has now implemented new protections to stop this behavior. Any blocked requests will be routed to its older Opus 4.8 model.
Also read: Chinese AI firm Z.ai takes aim at Anthropic’s Cloud Myth
Anthropic says that while this may be frustrating for users, the trade-off is to ensure that Fable 5’s other features remain widely available. The company also warned that it “may not be possible” to make any AI model fully resistant to “jailbreaking,” a term used to refer to techniques for bypassing security measures.
“There will be many minor jailbreaks, some of which are narrowly harmful, and although no universal jailbreak for Fable 5 has been discovered at the time of writing, expert security researchers continue to red team it,” the company said.
Close coordination, supervision and its implications
Anthropic also said in a blog post that it is deepening its collaboration with the U.S. government, allowing select government partners to expand early access to both of its models.
U.S. Commerce Secretary Howard Lutnick said in a letter to Anthropic that the company has agreed to work with the government on agreements for Mythos, Fable and future models and to notify the U.S. government of any malicious activity. However, Lutnick said the department “reserves the right to reevaluate the decisions made in this letter and the need to reimplement permitting requirements” if circumstances change or Anthropic fails to comply with its commitments.
Increase scrutiny of AI models In June, US President Donald Trump signed an executive order establishing a voluntary framework for AI developers to provide the government with “covered cutting-edge models” for 30 days before releasing them to trusted partners.
Anthro’s rival OpenAI also faces limitations. It said last week that it was delaying the full public release of GPT-5.6 at the request of the U.S. government, limiting its access to a small group of vetted partners.
The incident marks a shift in how Washington now views AI policy. Traditionally, the U.S. has imposed export controls on the hardware that powers these models, rather than the finished product and publicly deployed software. The government’s brief withdrawal from consumer products suggests that such intervention may become the new normal in a post-AI world.
Also read: Why Anthropic has an advantage over OpenAI in the IPO race
Isaac Harris, executive director of the Frontier Security Institute, a nonprofit focused on artificial intelligence and national security, told Reuters there now appears to be a standard U.S.-model process. But he added, “There remains a question mark over how the U.S. government will deal with an equally dangerous capability coming from China with fewer safeguards.”
Humanity vs Pentagon
This closure is the latest for the United States Government The feud has been going on for months.
Dean Ball, the former artificial intelligence adviser to the Trump White House and the main drafter of the government’s 2025 artificial intelligence action plan, once said on the New York Times podcast that Anthropic’s relationship with the US government can be traced back to the summer of 2024.
Power said on “The Ezra Klein Show” that at the time, the Joe Biden-era Pentagon and Anthropic had agreed to use Crowder in classified settings, including intelligence analysis, with two exceptions: no domestic mass surveillance and no use of fully autonomous lethal weapons.
The Trump administration expanded the contract with the same terms in 2025, but relations soured throughout the fall and became public in early 2026, Ball said.
He said the decisive break came when Emil Michael, the undersecretary of war for research and engineering, asked Anthropic to remove a clause prohibiting the use of Cloud to analyze large amounts of collected commercial data, a capability at the heart of the mass surveillance objections.
The fight is also incredibly personal and political, Ball noted.
U.S. Defense Secretary Pete Hegseth, writing in X, said Anthropic’s “real goal is unmistakable. To seize veto power over U.S. military operational decisions,” which was “unacceptable.” Trump also called Anthropic a “radical left-wing, woke company” and its employees as “left-wing nut jobs.”
Things came to a head in February, when Hegseth threatened to invoke the Defense Production Act and designate Anthropic as a “supply chain risk.”
The label is typically reserved for technology deemed too dangerous to appear anywhere in the U.S. military supply chain, and has been used to target foreign companies such as China’s Huawei over espionage concerns. Never before has an American company been labeled as such.
Trump later announced on the Truth social networking site Every federal agency will stop using Anthropic technology: “We don’t need it, we don’t want it, and we won’t do business with them anymore!” The US president also warned the company to “help during this phase-out period, or I will use the full power of the President to make them comply, with significant civil and criminal consequences.”
Anthropic, whose Pentagon work involves $200 million worth of contracts, said being designated a supply chain risk “is not only legally unsound but would set a dangerous precedent,” adding: “No amount of intimidation or punishment from the War Department will change our position on large-scale domestic surveillance or fully autonomous weapons.”
The Department of Defense eventually signed a deal with OpenAI, which said it had the same red lines as Anthropic. OpenAI chief executive Sam Altman told staff in a note seen by the BBC that his company would reject military uses “such as domestic surveillance and autonomous offensive weapons”.
Anthropic subsequently filed two federal lawsuits against the Department of Defense on March 9, alleging that the “supply chain risk” label was improperly used and amounted to retaliation for the company’s protected speech on artificial intelligence safety and violated its due process rights.
“The Constitution does not allow the government to use its vast power to punish companies for protected speech. No federal statute authorizes the actions taken here,” Anthropic argued. He told the court that the designation would cause “irreparable harm” and put “hundreds of millions, if not billions, of dollars in lost revenue” at risk.
The case remains unsolved. A federal judge in San Francisco issued a preliminary injunction against Claude, but the DC Court of Appeals denied Anthropic’s request for an emergency stay on April 8.
(With input from various agencies; this article is an update of an article published in June)
