Just days after releasing its latest artificial intelligence model, Fable, Anthropic was compelled by the Trump administration to take the system offline under tight and unprecedented government restrictions. The directive came shortly after the company's launch, with officials giving Anthropic a mere 90 minutes to comply. By late Friday, a legal order was issued barring all foreign nationals—including Anthropic’s own non-U.S. employees—from accessing the Fable system, effectively forcing the company to suspend use of both Fable and its related model, Mythos.

The move marks a significant escalation in the ongoing tension between the White House and the San Francisco-based AI firm. According to officials familiar with the discussions, the government acted after warnings from several companies, including Amazon, that the new model could be vulnerable to misuse by hackers. Despite Anthropic’s assurances that Fable posed no greater risk than similar AI products from competitors such as OpenAI, concerns about the model’s potential exploitation spurred the sudden intervention.

Anthropic and the Trump administration have clashed before, particularly over how the company’s technology might be used by the military. Some Silicon Valley insiders sympathetic to the White House have suggested Anthropic has lobbied for regulatory constraints that could disadvantage smaller AI firms. Still, Anthropic remains a rising powerhouse in the AI industry, recently surpassing OpenAI in valuation and preparing for a public offering that could significantly enrich its founders and employees.

The restrictions were imposed using export control authorities typically reserved for sensitive technologies with dual civilian and military applications. Experts describing the use of a formal government demand, known as an “is informed” letter, characterized it as a rare and forceful step to address national security concerns. Penalties for ignoring such an order include severe corporate and personal criminal liabilities.

The government’s concerns center on fears that Fable’s built-in safeguards against malicious use, such as computer hacking, could be bypassed—a practice known as “jailbreaking.” Amazon CEO Andy Jassy personally raised these fears, and some officials claimed Anthropic was initially dismissive of the concerns and reluctant to cooperate. However, representatives close to Anthropic counter that the company was willing to address viable issues but was not provided with concrete evidence beyond Amazon’s report.

Anthropic had originally withheld Mythos from public release due to its potent hacking capabilities, a factor that contributed to the government’s scramble to develop oversight policies. The White House had established a voluntary system requiring AI developers to share forthcoming models with national security officials in advance, but some insiders now question the efficacy of this framework, citing the latest events as evidence of its limitations.

Over the weekend, government officials and allied voices renewed criticism of Anthropic’s approach on social media platforms. Defense Secretary Pete Hegseth, for example, expressed support for removing Anthropic’s technology from Pentagon systems months earlier.

Anthropic maintains that no AI safeguards are foolproof but insists that the risks triggering the government’s actions were overstated. The company said it collaborated extensively with U.S. and British authorities and independent experts to test Fable’s protective measures before release.

Ongoing talks between Anthropic’s technical staff and White House officials continued through the weekend, as both sides seek to resolve the dispute. Security experts have noted that alleged “jailbreak” exploits rely on multiple user prompts that may be typical defensive tactics, complicating attempts to simply “patch” the model without impairing legitimate uses.

The standoff underscores growing tensions as U.S. authorities contend with regulating sophisticated AI technologies that have potentially wide-reaching national security implications, even as companies race to commercialize advanced models in a highly competitive sector.