Tensions between the artificial intelligence company Anthropic and the Trump administration resurfaced over the weekend after the U.S. government directed Anthropic to block foreign access to its most advanced AI models. The order, aimed at addressing national security concerns, presents a challenge to Anthropic’s business operations and its reputation for prioritizing AI safety.
Anthropic complied by disabling access to its top-tier systems, including the recently released models Fable 5 and Mythos 5, for all customers outside the United States. However, the company continues to provide access to other, less advanced models. Fable 5, known for excelling in software engineering and scientific research, and Mythos 5, a tool designed to identify and rectify software vulnerabilities, were both released just days prior to the government’s directive. The company had initially limited Mythos 5’s availability to a select group of cybersecurity professionals and infrastructure providers due to concerns about potential exploitation by hackers.
The U.S. government cited undisclosed national security risks and indicated it had identified a way to circumvent the safeguards—sometimes referred to as “jailbreaking”—embedded in Fable 5 to prevent misuse. Anthropic, however, maintained that no tester had successfully bypassed these protections in a broad manner and that it had collaborated with government officials before releasing the models.
David Sacks, a former Trump administration crypto and AI advisor and current venture capitalist, criticized Anthropic for prioritizing consumer access over security. He stated that the administration seeks to resolve safety issues promptly so that access to the model can be restored, expressing frustration over what he described as Anthropic’s reluctance to fully comply with safety demands.
The dispute intensified on social media, with Trump officials renewing attacks on Anthropic. Defense Secretary Pete Hegseth recalled that the Department of War had expelled Anthropic from its facilities three months earlier and suggested that recent developments vindicate that decision.
The government’s intervention reportedly stemmed in part from conversations involving Amazon CEO Andy Jassy and senior Trump administration officials, including Treasury Secretary Scott Bessent. Amazon researchers are said to have demonstrated that Fable 5 could be manipulated to provide information useful for cyberattacks. Amazon, which has invested billions in Anthropic and supplies chips for its AI training, confirmed that it routinely advises governments on security risks but did not comment directly on the matter.
Anthropic disputed that these findings warranted a broad restriction on its commercial model, arguing that applying such a standard across the AI industry would stall innovation among all leading developers.
This episode marks another chapter in a fraught relationship between Anthropic and the Trump administration. Previously, the two parties clashed over a $200 million Pentagon contract, with Anthropic seeking assurances that its technology would not be used for surveillance of U.S. citizens or autonomous weapons—a stance the Defense Department saw as an attempt to exert undue control over military operations. In February, former President Trump instructed federal agencies to cease using Anthropic’s technology, labeling the company as “radical left” and “woke,” prompting Anthropic to sue the administration for retaliatory measures and challenge its designation as a supply chain risk.
Efforts are reportedly underway to address the government’s security concerns. Senior Anthropic technical staff traveled to Washington on Monday to meet with White House officials at the Commerce Department, a source said, though the department had not commented publicly on the meeting. The outcome of these talks may determine the future availability of Anthropic’s most powerful AI models both domestically and internationally.