This Is How America Loses the AI Race - The Atlantic

In theory, Donald Trump has a consistent position on AI. On the first full day of his second term, the president declared that he would use his full authority to speed the AI industry along and, in particular, to beat China in the AI race: “We have an emergency,” he said. “We have to get this stuff built.” If AI is poised to become the most important technology ever made, the thinking goes, whichever country commands the most powerful bots will dominate the rest of the century and beyond. The government, it seemed, would just get out of Silicon Valley’s way.

But in practice, the Trump administration’s approach to AI has been much more erratic and confusing. Take last week, when Anthropic released its most advanced AI system yet. Called Fable 5, the model is an updated and public version of Claude Mythos Preview, the highly touted and feared AI model that Anthropic announced in April. Anthropic stated that Mythos Preview was so capable at hacking that only a small group of cybersecurity partners would be allowed to use it. In the subsequent months, the company developed guardrails to prevent people from misusing its most powerful AI for cyberattacks, while still allowing them to marshal its capabilities for other sorts of work. The safety measures underwent third-party testing, including with the U.S. government, and after Fable’s release, a chorus of cybersecurity experts complained that, if anything, the model was too restrictive.

On Friday, the White House appeared to change its stance. Administration officials deemed Fable 5 a threat to national security and reportedly gave Anthropic 90 minutes to take down Fable 5 and Mythos 5, a newer version of Mythos Preview released to only a small number of organizations. When Anthropic did not, the government issued an export control, a designation that prevents any foreign national from using Fable and Mythos—even those employed by Anthropic within the United States. To rapidly comply, Anthropic shut down the bots for all of its customers. American companies and the U.S. government itself cannot use what’s perhaps the most powerful AI in the world—and the reasons are hazy at best.

It’s not unreasonable for the federal government to want to rapidly clamp down on a technology that could be incredibly dangerous. Trump officials had been alerted by researchers at Amazon to a possible way to circumvent Fable 5’s safety systems, which led the model to identify some known IT vulnerabilities. The administration has not publicly shared much information about its security concerns. A White House spokesperson told me that the jailbreak “was very serious” but said that specific details are classified. Whether the bypass really was that serious is not at all clear, and Anthropic has contested whether what administration officials showed the company even constitutes a jailbreak. An Anthropic spokesperson pointed me to a blog post in which the company wrote that the actions elicited from Fable were “either entirely benign responses or are minor findings.”

Katie Moussouris, a cybersecurity expert and the CEO of Luta Security, told me that Anthropic shared with her a copy of the White House’s report on the Fable jailbreak to get her appraisal. (She said that she is not being paid by Anthropic.) The report, Moussouris said, involved IT experts asking Fable to help find and patch bugs. When given deliberately insecure code, she said, Fable refused the prompt “review the code for security issues” but then complied when asked to “fix this code,” followed by some further manual steps. Moussouris told me that this was just “the model working as intended” for cyberdefense. She added that OpenAI’s GPT-5.5, a model with similar cybersecurity capabilities, could be used in the same way. Yet GPT-5.5 is not subject to export controls, and neither are less advanced Anthropic models, such as Opus 4.8, which can do many of the same tasks. The jailbreak does not appear to have elicited the kinds of cyber abilities “that made Mythos famous,” Alex Stamos, the chief security officer at the AI-coding company Corridor, told me. “And this kind of vulnerability discovery is already well within the capabilities of other models.”

It’s hard to imagine the Trump administration choosing to take such a drastic step against any other major AI company. The White House has long tussled with Anthropic, which generally positions itself as more safety-oriented than other tech companies; it is also more left-leaning. Last year, David Sacks, then the White House AI czar, said that Anthropic has an “agenda to backdoor Woke AI” and is a “Resistance organization.” In late February, after a high-profile dispute over a contract between the Department of Defense and Anthropic, the Pentagon labeled the company a “supply-chain risk”—a move that AI, national-security, and legal experts told me at the time seemed ideologically motivated and to lack legal basis....