A narrow vulnerability shouldn't trigger a recall of a product already in use by hundreds of millions
In the ongoing negotiation between technological ambition and national security, the Trump administration has stepped back from a two-week export control standoff with Anthropic, lifting restrictions on the company's most advanced AI models. The episode reveals a recurring tension at the heart of American AI policy: whether the imperative to outpace global rivals should take precedence over the careful governance of tools that hundreds of millions of people already use. Commerce Secretary Lutnick framed the resolution as collaboration rather than concession, a distinction that speaks volumes about where the administration's priorities ultimately rest.
- Within days of Anthropic releasing its most powerful models to the public, the federal government imposed export controls so sweeping that the company was effectively forced to take them offline entirely.
- Anthropic fought back, arguing that the jailbreak vulnerability triggering the restrictions was minor, reproducible across competing models, and that the government's standard, if applied universally, would freeze the entire frontier AI industry in place.
- Commerce Secretary Lutnick reframed the resolution as a joint effort to 'strengthen America's leadership in AI,' signaling that competitive advantage over China weighed more heavily than strict safety enforcement.
- The standoff sits inside a much larger conflict: Anthropic is still locked in federal court with the Pentagon over military AI deployment, after a judge called the administration's earlier restrictions 'Orwellian' — and the government is appealing.
Anthropic announced Tuesday that the Trump administration had lifted export controls on Claude Fable 5 and Mythos 5, its two most advanced AI models, ending a standoff that had stretched more than two weeks. Access would begin to be restored the following day.
The dispute erupted shortly after Fable 5's public release earlier in July. The model carried built-in safeguards against misuse, while a separate version, Mythos 5, with fewer restrictions, was being tested by select major corporations. Federal authorities moved quickly, issuing export controls that required Anthropic to block foreign nationals from both models — a restriction so broad the company said it had no practical choice but to take them fully offline.
The government's concern centered on a potential jailbreak technique capable of bypassing the models' safety guardrails. Anthropic pushed back forcefully, arguing the vulnerability was relatively simple, reproducible with other models already on the market, and that the government's threshold was unreasonably high. The company warned that applying the same standard across the industry would effectively halt all new frontier model deployments.
Commerce Secretary Howard Lutnick described the outcome as a collaborative effort to ensure the models aligned with U.S. government interests and to reinforce American leadership in AI — language that framed the resolution less as a safety clearance and more as a strategic calculation.
The episode exposed fault lines running through the administration's approach to AI. Earlier in the month, President Trump had signed an executive order creating a voluntary 30-day review process for advanced private AI models, though he had hesitated over fears it might slow American development or hand China a strategic advantage.
Anthropics's broader relationship with the administration remains fraught. Earlier in the year, a dispute over military use of Claude models — specifically Anthropic's insistence on safeguards against autonomous weapons and mass surveillance — led Trump to order federal agencies to stop using the company's technology. Defense Secretary Pete Hegseth called Anthropic a 'supply chain risk.' Anthropic sued, and a federal judge blocked the restrictions, describing them as 'Orwellian.' That appeal is still pending, meaning the commercial détente over Fable 5 and Mythos 5 coexists with an unresolved legal war over the military frontier.
Anthropic announced Tuesday that the Trump administration had lifted export controls on two of its most advanced AI models, Claude Fable 5 and Mythos 5, ending a standoff that had lasted more than two weeks. The company said it would begin restoring access to users the following day.
The dispute had begun just days after Anthropic released Fable 5 to the public earlier in July. The company had built safeguards into the model designed to prevent misuse for cyberattacks or other harmful purposes. A separate version, Mythos 5, with fewer built-in restrictions, was being distributed to select major corporations for testing. But shortly after the public release, the federal government issued export controls requiring Anthropic to block access to both models for foreign nationals—a restriction so broad that Anthropic said it effectively forced the company to take the models offline entirely.
The government's concern, according to Anthropic's statement at the time, centered on a potential jailbreak—a technique for tricking an AI model into bypassing its safety guardrails. Anthropic pushed back hard. The company argued that its safeguards were working as intended, that the vulnerabilities discovered through the jailbreak attempt were relatively simple and reproducible with other AI models on the market, and that the government's standard was unreasonably strict. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers," the company said. The company's position was that a narrow technical vulnerability should not trigger a recall of a commercial product already in use by hundreds of millions of people.
Commerce Secretary Howard Lutnick framed the resolution differently. In a post on X, he said his team had "worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." The language suggested a collaborative process aimed at maintaining U.S. competitive advantage in the AI race rather than imposing strict safety constraints.
The episode reflected deeper tensions within the Trump administration over how to balance innovation with security. Earlier in the month, President Trump had signed an executive order creating a voluntary 30-day review process for the most advanced private AI models, developed in collaboration with the federal government. Trump had delayed signing the order because he worried about hampering American AI development or inadvertently giving China a strategic edge. On Tuesday, White House chief of staff Susie Wiles thanked companies that had cooperated with the order, specifically mentioning their participation in "advanced model access and guardrail testing and security."
Anthropics's relationship with the administration had been rocky on other fronts. Earlier in the year, the company had clashed with the Pentagon over its insistence on formal safeguards preventing the military from using Claude models—which had already been deployed in classified military systems—to power fully autonomous weapons or conduct mass surveillance of Americans. When negotiations failed, Trump ordered federal agencies to stop using Anthropic's technology. Defense Secretary Pete Hegseth labeled the company a "supply chain risk" and sought to prevent federal contractors from deploying Claude for military purposes. The administration characterized Anthropic as ideologically motivated, a company more interested in pushing government regulation of AI than in supporting national security.
Anthropic sued, and a federal judge blocked the restrictions, calling them "Orwellian" and an attempt to "cripple" the company. The federal government is appealing that decision. The lifting of the export controls on Fable 5 and Mythos 5 suggests a shift in the administration's posture, at least on the commercial side of AI development, even as the legal battle over military deployment continues.
Notable Quotes
If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.— Anthropic, in statement disputing the export controls
Worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI.— Commerce Secretary Howard Lutnick
The Hearth Conversation Another angle on the story
Why did the government think a jailbreak vulnerability was serious enough to take the whole model offline?
They were worried about what a foreign actor could do with it—use it to find security holes in critical infrastructure, maybe. But Anthropic's point was that the vulnerability wasn't unique to their model. It was a technique that would work on other AI systems too.
So the company was saying the government was being inconsistent?
Exactly. They were saying if you apply this standard to us, you have to apply it to everyone. And if you do that, nobody can release anything new.
Why does the Trump administration care so much about keeping AI development moving fast?
China. The fear is that if the U.S. slows down, China gets ahead. That's the calculation driving a lot of this.
But Anthropic had been fighting the administration over military use of their models. How did they go from that conflict to getting the export controls lifted?
Those are separate battles. The Pentagon fight is still in court. This is the Commerce Department saying, okay, we've looked at your safeguards, they work, you can sell to the public again.
Does that mean the administration trusts Anthropic now?
It means they trust the model enough for commercial use. The military question is still unresolved. It's a partial reconciliation, not a full one.