The Trump administration has imposed export controls prohibiting foreign national access to Anthropic’s most advanced artificial intelligence models, Mythos and Fable, citing national security concerns. The directive, which emerged late last week, marks an unprecedented move by the U.S. government to restrict foreign use of software AI developed by an American company.

The restrictions followed reports from Amazon’s CEO Andy Jassy, who alerted administration officials that Amazon researchers had successfully “jailbroken” Fable 5—bypassing its safety mechanisms in a manner that could potentially aid cyberattacks. According to sources familiar with the matter, Amazon shared its findings with the government, prompting an expedited discussion involving Treasury Secretary Scott Bessent, White House Cyber Director Sean Cairncross, and Commerce Secretary Howard Lutnick.

Following the government’s order, Anthropic responded by disabling access to Mythos and Fable entirely for all users to comply swiftly. The company explained that some of its own employees are foreign nationals, making compliance with the export controls complex. Anthropic described the reported vulnerability as a “narrow potential jailbreak” and disagreed with the government’s decision to broadly restrict access, arguing that the flaw was not unique to Fable and that applying such a standard industry-wide could halt releases from all frontier AI providers.

The government’s decision reflects broader concerns about AI safety risks, especially after reports surfaced that a China-linked group had possibly clandestinely accessed Mythos, though Anthropic already prohibits access to its products within China. While the White House emphasized the move as necessary for national security, some officials criticized Anthropic for downplaying the concerns and not acting promptly to address or pause external access once vulnerabilities were identified.

Several senior Trump administration officials expressed distrust towards Anthropic’s handling of the situation, pointing to previous misgivings about the company’s approach to cybersecurity. One official remarked that Anthropic had publicly raised alarms about AI risks while seeming reluctant to treat its own models with equal caution. Another noted that multiple companies, not just Amazon, had conveyed worries about Fable prior to the administration’s intervention.

Anthropic CEO Dario Amodei has openly warned about the dangers of advanced AI, including predictions that it could cause severe economic disruption or even pose existential risks. Yet, some administration figures criticized the company’s contradictory stance, noting that despite its dire warnings, Anthropic was viewed as not sufficiently cooperative when asked to mitigate potential threats.

The company is currently working intensively to address regulatory concerns in hopes of restoring access to its models. Treasury Secretary Bessent and White House Cyber Director Cairncross have been involved in ongoing discussions with Anthropic leadership. Meanwhile, AI industry observers warned that the administration’s unprecedented move could set a precedent affecting other major AI developers and the future of AI deployment and regulation.

Anthropic, founded by former OpenAI staff including Amodei in 2021 with a focus on safety-conscious AI development, had previously limited Mythos access to about 40 top corporations, including Apple and JPMorgan. Despite these precautions, Mythos’s capability to uncover cybersecurity vulnerabilities led to caution within the White House and calls for tighter oversight.

The ongoing dispute between Anthropic and the Trump administration also reflects deeper tensions. The Pentagon recently designated Anthropic a supply-chain security risk, prompting the company to challenge the decision legally. Additionally, Amodei has criticized some Trump policies and reportedly hired former Biden administration officials, further complicating the relationship.

While the administration’s intervention underscores growing governmental control over powerful AI technologies, it could impact Anthropic’s market position, especially as it prepares for a potential initial public offering amid increasing competition from rivals like OpenAI and Google. The government’s stated aim remains ensuring safety without hindering innovation, but the situation highlights challenges in balancing rapid AI development with national security concerns.