The White House vs. Anthropic's New AI Model

Is the Trump administration just hellbent on thwarting Anthropic now? Citing "national security" concerns, it has effectively killed public access to Anthropic's latest artificial intelligence model. The move sure looks like payback for Anthropic's earlier refusal to grant the government the right to use its products for mass surveillance and robot weapons.

Antrhopic's newest public-facing model is called Claude Fable 5. Launched on June 9, it was billed as a "safe for general use" version of Claude Mythos 5, Anthropic's powerful new AI model that was released only to "a small group of cyberdefenders and infrastructure providers…in collaboration with the US government."

The White House told Anthropic in a Friday letter that under a new export control directive, Mythos and Fable 5 could not be used by any foreign national in the U.S.—including Anthropic employees here on work visas—or anyone in a foreign country.

With no way to immediately comply with that, Anthropic temporarily ended public access for everyone.

Does Fable Pose a National Security Threat?

The government cited national security as the reason for its new directive, but its letter "did not provide specific details of its national security concern," noted Anthropic on Friday.

"Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5," Anthropic continued.

This jibes with what Axios has reported: that the White House acted "after an urgent report from Amazon" that "they were able to jailbreak and access portions of Anthropic's powerful new Mythos model that pose a national security threat."

But according to Anthropic, the hubub comes down to "a small number of previously known, minor vulnerabilities." These "all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass." The company has "not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift," it said.

To comply with the Trump administration's new directive, Anthropic blocked all access to Fable 5 and Mythos 5 for now.

It says it is "working to restore access as soon as possible."

Is This Retaliation?

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic stated in its Friday post. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

"We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts," it added. "This action does not adhere to those principles."

Anthropic didn't outright accuse the administration of retaliation.

Extent to which White House allies are signaling that this is a culture war issue, not a technical one, is striking https://t.co/1fgoJEJn64

— Ben Smith (@semaforben) June 13, 2026

But Anthropic has been in the White House's crosshairs for months, since the company—which had a contract with the Department of Defense—refused to remove limits on using Claude for mass surveillance or autonomous weapons. In response to this refusal, the administration designated Anthropic a "supply chain risk" and forbade any government agency or contractor from using an Antrhopic technology.

Anthropic fought back with a lawsuit accusing the administration of illegal retaliation and of "wield[ing] its enormous power to punish a company for its protected speech."

U.S. District Judge Rita Lin preliminarily sided with Anthropic, writing that the Pentagon had offered "no legitimate basis to infer" that Anthropic "might become a saboteur" and rebuffing "the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." The government may not commit "corporate murder" just because a company won't do whatever it wants, Lin wrote, issuing a temporary block on the supply chain risk designation.

Ultimately, the government was unable to bully Anthropic into changing its policies and unable to frighten Anthropic into remaining quiet about what happened. And so far, it's been unable to punish Anthropic by blocking people from doing business with it.

In the wake of this, it suddenly realizes that Anthropic's new model is a threat? It seems suspicious. Especially when you look at the way officials like Defense Secretary Pete Hegseth are talking about the current situation.

Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever.

Every passing day proves why that was the right move.

— Pete Hegseth (@PeteHegseth) June 13, 2026

'Way Out of Line'

Before releasing Fable 5, "Anthropic had previously notified the government multiple times…and the government did not object," according to Axios. 

Luta Security CEO Katie Moussouris told Axios that the government's response "seems way out of line with what's actually in the [Amazon] research report" and that the discovery of the security vulnerabilities actually demonstrates things working correctly. "All AI models need to be able to help [security] defenders in exactly this way, or we won't be able to scale our defense against attackers," Moussouris said.

It's also worth considering the way the Trump administration responded. If the vulnerabilities in question are really so concerning, wouldn't they be concerning in the hands of Americans too?

These vulnerabilities are supposedly so dangerous that we must block all foreigners—not just from hostile countries, but allied countries too—from access, but they're also perfectly OK for American citizens to use. The idea seems a bit suspect. It suggests that this is more about security theater and/or coming up with a plausible pretense for punishing a company that pissed off Trump-world officials than addressing legitimate concerns.

Not everyone agrees, of course. David Sacks, the White House's former AI czar, has said this isn't about the prior disagreement and only comes down to preventing people from accessing a "cyber weapon."

Regardless of whether the administration's actions are a deliberate attempt to harm Anthropic, they certainly signal an aggressive approach to AI regulation—and a departure from the rhetoric in President Donald Trump's recent executive order.

That order explicitly rejected the "creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models." But this rejection means nothing if models that don't get preclearance can suddenly find themselves facing severe restrictions after they're deployed anytime authorities are feeling spooked.

Be Careful What Your Wish For

"Unlike the government's legally dubious attempt to brand Anthropic a 'supply chain risk' earlier this year, there is at least a facially plausible legal framework in this case," notes University of Minnesota law professor Alan Rozenshtein at Lawfare. The administration is likely relying on the Export Control Reform Act of 2018, which covers technology and "treats releasing controlled technology to a foreign person inside the United States as an export to that person's home country."

But "the current situation thus marks the first time that export controls have been enforced to control access to an AI model," Rozenshtein observes. If "Washington is going to treat frontier models as national-security assets it can switch on and off," he suggests, then Congress needs to set some guidelines around that.

Perhaps. But it seems naive to imagine that there's some sort of objective and neutral way that Congress could direct the authorities here. If the president or the Pentagon gets the right to unilaterally block the release of new AI models, deem them unfit for foreign nationals, or otherwise have a tight grip on AI output, that's always going to lend itself to the sort of "political favoritism or arbitrary decisions" that Anthropic CEO Dario Amodei wrote of earlier this month.

Amodei is among the tech leaders who have actually promoted greater regulation of AI.

— hoopy frood (@huwupy.kawaii.social) 2026-06-13T14:58:17.019Z

"Because this is happening to Anthropic, the temptation for many will be to say: Play stupid games, win stupid prizes," writes Adam Thierer of the R Street Institute. "They have relentlessly raised the regulatory temperature in Washington by inviting far-reaching controls of frontier models….

"But this decision by the Trump administration should not be judged on a desire for payback politics, but on the merits, and specifically what it means for America's broader AI objectives," Thierer adds. "In that regard, this action is truly outrageous. How exactly is the government planning on even going about verifying everyone who uses this specific model to ensure compliance? That alone raises huge flags." 

In the News

"Taken together, these bills would fundamentally change the internet as we know it." A new tech policy package being negotiated between Congress and the White House looks like very bad news for free speech and privacy online. "In exchange for overriding some state AI laws, the deal would reportedly involve a sweeping package of federal restrictions on online speech—paving the way for an unprecedented era of online censorship," writes the Foundation for Individual Rights and Expression. "The legislative package could include the Senate version of the Kids Online Safety Act, the NO FAKES Act, and age verification requirements."

Just so we're clear, trading preemption of some state AI laws for federal bills like KOSA, NO FAKES, GUARD Act, or ASAA is a bad deal for free speech online.

— John Coleman (@FreeSpeech_AI) June 12, 2026

U.K. to ban teens from social media, online gaming. The U.K. says it will ban under-16-year-olds from social media—a move that will require checking IDs or some other form of invasive identify verification for all users. "The government said it also planned to block children under 16 from livestreaming and from communicating with strangers on a wider range of online services, including game sites," The New York Times reports.

Australia did this late last year. It hasn't been working.

"Six months in, most indications are that the law has largely failed at keeping young teens off the platforms, in a disappointing start to an initiative carefully watched by parents and governments around the world," writes Times Australia correspondent Victoria Kim.

A recent report from Australia's government pointed out that 70 percent of parents say their kids under age 16 still had social media accounts. In another poll, "three in five (61%) Australian 12-15 year-olds who had accounts on restricted platforms before the ban came into force still have access to one or more account."

But when have regulators and government officials ever been deterred by clear evidence of failure? Full steam ahead!

On Substack

AI chatbots could save lives. Artificial intelligence can be beneficial in certain medical contexts, detecting diseases earlier and possibly saving lives. That includes those much-maligned AI chatbots, suggest Adam Omary and Jennifer Huddleston at Human Progress:

The Senate Judiciary Committee advanced Senator Josh Hawley's Guidelines for User Age-verification and Responsible Dialogue (GUARD) Act. The bill would require every American to verify their age before using a generative AI chatbot and would bar anyone under eighteen from using a "companion" chatbot at all. In the room during the markup were the parents of children who died by suicide after conversations with AI products. Their grief is unimaginable, and their motives are beyond reproach. But concerningly, such a policy might quietly cost rather than save lives.

The strongest claim animating this bill is the belief that restricting minors' access to AI chatbots will prevent suicide. On the available evidence, that claim is closer to a hypothesis than a finding—and a hypothesis that runs against several decades of data on how young people die.

According to the Centers for Disease Control and Prevention, the American suicide rate began climbing around the year 2000—before ChatGPT, smartphones, or social media even existed. It accelerated through the 2010s, then, contrary to popular narrative, plateaued and modestly declined after 2018—even as generative AI moved from research labs into the pockets of nearly every teenager in the country. If chatbots were a meaningful driver of adolescent suicide, the curves should have moved together. They have not, and, importantly, suicide rates among young Americans remain the lowest among any age group.

Omary and Huddleston go on to suggest that for some adolescents—especial those "in households where therapy is unaffordable, unavailable, or unsafe to disclose"—chatbots could be an accessible and "reliable form of emotional support" and actually help prevent self harm and suicide.

Read This Thread

This is actually a perfect example of "what's the difference between a bouncer checking your ID and a website doing it?"

Identity theft. https://t.co/c6S6Oveih2

— Shoshana Weissmann, Sloth Committee Chair (@senatorshoshana) June 12, 2026

More Sex & Tech News

• The Justice Against Weaponized Bureaucratic Overreach to Networked Expression (JAWBONE) Act, from Sens. Ron Wyden (D–Ore.) and Ted Cruz (R–Texas), aims to curtail indirect government censorship of speech on social media, AI systems, and TV.

• A federal appeals court is considering the constitutionality of an Arkansas law that criminalizes librarians who provide books that authorities consider "harmful to minors" and give city and county officials the power to determine library content.

• Nice to see people pushing back against unfounded anti-porn tropes.

• An Ohio police chief is facing 70 counts of felony charges related to sex crimes. Bethel police chief Chad Essert is accused of sexual battery and unlawful sexual conduct with a minor for alleged offenses that took place from 2005–2010. Essert also faces a sexual harassment claim from a former employee and was under investigation for having sex on duty, reports WKRC.

• The American Innovation and Choice Online Act is still bad. (More on an earlier version of this bill here.)

• Self-help author Tim Ferris suggests AI is already having a dire effect on how-to nonfiction sales…but comes away with a surprisingly/refreshingly optimistic take on the phenomenon.

• Illinois is issuing a special monthly tax on social media platforms. "The proposal's biggest hurdle is the decades of case law that have squarely labeled this kind of tax as exactly what it is: a regulation of speech," writes Tyler Tone.

• Ahead of the World Cup in Los Angeles "vast resources are being put towards crackdowns and sweeps," writes Antonia Crane. "Sex workers—like me—get arrested because cops and politicians have something to prove."

The post The White House vs. Anthropic's New AI Model appeared first on Reason.com.