Late on Friday, Anthropic issued a surprise announcement: It was disabling its new Fable 5 model, released just days earlier, after a directive from the U.S. government that cited national security concerns. The model is a safeguarded version of Anthropic’s Mythos family of models that were deemed so powerful when first unveiled in April that it caused widespread alarm over cybersecurity threats.
In a June 12 blog post, Anthropic said it believes the government became aware of a “jailbreaking” method to bypass Fable 5’s safeguards. While the company is complying with the directive, it pushed back against it. “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,” the blog post says. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Just hours before the directive — Anthropic said it received the government’s letter at 5:21 p.m. Eastern Time on Friday — Chief Commercial Officer Paul Smith discussed the “tradeoffs” when it comes to releasing such a powerful tool in an interview with Forbes.
“You have to make a judgment call on these things,” he said. “The safest you can be is to not let people use something. And then it’s totally safe. But then how is that helping the mission, and how is that helping people actually derive real value from Mythos-level intelligence?”
Of course, he wasn’t talking about the directive, which hadn’t been issued yet. Instead, Smith’s remarks were in response to complaints by some users that Anthropic’s safeguards went too far, hamstringing the model with so many limitations, particularly with prompts related to biology (which could be used for bioweapons) and cybersecurity, that it rendered the model neutered. “Over the next few days and few weeks, we can tune and dial back those classifiers to make it a more smooth and seamless experience for users, but that was kind of like the tightrope we’re trying to thread,” he continued.
Still, Smith’s sentiment is a helpful lens through which to view Anthropic’s latest dustup with the government, which first sought to ban the use of Anthropic products for federal use in February and on Friday banned the use of Mythos by foreign nationals: Anthropic’s goal is to get the tools out there, and do it as safely as possible, but acknowledges that there are always risks.
White House AI czar David Sacks doesn’t agree. “In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety,” he wrote on X on Saturday.
Anthropic declined to comment on the situation beyond its blog post.
Meanwhile, the decision to disable the model has sent waves throughout the industry. It came as a surprise to some Anthropic customers. “Sounds like we’re going to have [to] turn off access to Fable,” Amjad Masad, CEO of Replit, an AI coding company that offers its users access to Anthropic’s models, posted on X late on Friday. According to a Reuters report that cited a source close to the matter, several tech executives including Amazon’s Andy Jassy had raised concerns to Trump over security risks tied to Anthropic’s latest models.
In March, CEO Dario Amodei engaged in a high-profile standoff with the Department of War over the Pentagon’s ability to use Anthropic’s AI, particularly regarding mass surveillance and autonomous weapons. After the tense back-and-forth, the government labeled Anthropic a “supply chain risk” — a devastating blow to any company that does business with the military. Anthropic subsequently sued the DoW, and a judge put an injunction on the designation.
Aside from the Fable safeguards, Smith touched on other topics during his interview with Forbes.
Asked if the DoW situation would have been more complicated as a public company, Smith demurred. “That’s a crystal ball thing,” he said. “That was a very complicated situation. I think that will always be a very complicated situation [public company or not].”
He declined to discuss a timeline for Anthropic’s planned IPO, after the company confidentiality filed its S-1 prospectus with the SEC last month. But he insisted that the company’s stated mission of AI safety wouldn’t be at odds with being a public company, when it would be subject to shareholder pressure. “There is no question in terms of Dario and the other co-founders’ commitment to safety, to trust, to keeping focused on that core mission,” he said. “Candidly, I think the number one question that’s going to be on a post IPO earnings call will be, how is AI diffusion going in the wider economy.”
But he did discuss Anthropic’s deal with Elon Musk’s SpaceX, which went public Friday in the biggest IPO in history, to run its models on SpaceX’s Colossus supercomputer. As part of the contract, Anthropic is paying SpaceX $15 billion a year through 2029, according to SpaceX’s S-1. “How it came about is, we always want to diversify our sources of compute. And this was just the perfect intersection of SpaceX also wanting to diversify who was using that compute,” Smith said. “It came together very, very, very, very quickly.”
Still, Anthropic and Musk make unlikely bedfellows. In February, he slammed the company on X, calling it “evil” and discriminatory. “Frankly, I don’t think there is anything you can do to escape the inevitable irony of Anthropic ending up being Misanthropic,” he wrote. “You were doomed to this fate when you chose your name.”
Smith brushes off those comments now. “I generally only react to what people say when they’re actually in a room with you and in real conversations, versus what is posted on X and other forums,” he said. “So rather, just deal with what’s in front of me.” He clarifies he wasn’t literally in the room when the deal was negotiated. That was Anthropic cofounder Tom Brown and CFO Krishna Rao, he says.
