No, the NSA wasn’t hacked by AI. Here’s what actually happened
Claims about artificial intelligence spread online like a forest fire, something Straight Arrow has previously reported on. Now, a new AI claim is circulating on social media following a recent Economist article about Anthropic’s Mythos ban.
The article details a conversation between Sen. Mark Warner, D-Va., the vice-chair of the Senate Intelligence Committee, and Gen. Joshua Rudd, the head of the Department of Defense’s Cyber Command. Warner said Rudd told him that Anthropic’s latest AI model was able to break “into almost all of our classified systems, not in weeks, but in hours.”
The quote spread quickly across social media, with some users believing Mythos had actually hacked into the National Security Agency. But it hadn’t. Shashank Joshi, the article’s author, pushed back on how his quote was being read.
“It would be a mistake to read that literally, I think,” Joshi wrote on X. “It surely depends on using Mythos alongside other tools under very particular conditions. I quoted it to give a sense of Mythos’ potency. But it was a mistake not to have added caveats.”
What actually happened?
Warner’s comments came from an internal test, not an outside intrusion, according to the International Business Times. Red teams ran the tests after Anthropic released Mythos early to major corporations and the government. Red-team exercises typically involve an organization testing its defenses by simulating an attack. In this case, the NSA reportedly used Mythos itself as the tool to probe its classified networks for weaknesses.
The NSA has not said what it found during testing, but Warner’s comments suggest Mythos accessed the NSA’s classified systems in far less time than it would’ve taken a team of people.
Anthropic released its Mythos model early for testing after saying it was capable of advanced cybersecurity tasks, such as detecting and creating exploits for critical software vulnerabilities, on its own.
Anthropic publicly released Fable 5, a consumer version of the same model, on June 9. Mythos itself is not available to the public — it remains limited to vetted partners, including the NSA, through Anthropic’s Project Glasswing program.
But three days later, the Trump administration issued an export-control directive to force the company to block the model’s use by foreign nationals, both inside and outside the U.S., including Anthropic’s own employees. The company said that to comply with the order, it would have to suspend access to the models to all customers.
How the quote escaped its context
Warner first made his comments during a hearing on June 11 while arguing for mandatory pre-release testing of cutting-edge AI models. During the hearing, he said Rudd told him Mythos “broke into almost all of our classified systems, not in weeks, but in hours.”
The Economist accurately quoted his statement, but over the weekend, users on social media stripped the quote from its context and rephrased it as “NSA confirms a breach,” which did not happen.
NewsGuard, a company specializing in media credibility, told Straight Arrow that oftentimes, alarming or absolute phrasing tends to outrun its context. Sofia Rubinson, NewsGuard’s reality check editor, said claims like this spread faster because of how they’re framed.
“When claims are made in definitive and dramatic terms, they are more likely to elicit an emotional reaction,” Rubinson told Straight Arrow. “These types of posts are designed to override our logical side and are oftentimes believed and shared without questioning the source.”
Rubinson praised Joshi for speaking up about how his quote was being misread. But she said clarifications like his rarely travel as far as the original claim did.
“It’s always great when people at the center of false claims publish a public clarification or correction,” she told Straight Arrow. “But due to the nature of social media, these corrections often do not go viral, and may fail to reach the people who originally fell for the false claim.”
Why does this matter beyond the tweet?
The new social media discourse about Anthropic comes as the company is in the middle of an export-control fight with the federal government. The legal basis for Mythos’ export controls is contested and unprecedented, as it’s the first time the U.S. has applied export controls directly to an AI model rather than to hardware.
The confusion has led to real costs, too. Allied nations are no longer allowed to use Mythos, including the United Kingdom’s AI Security Institute, one of the world’s leading bodies for testing cutting-edge AI models.
Meanwhile, Anthropic has finished testing its successor to Mythos, a sign that this fight isn’t slowing down anytime soon.
As the battle between the Trump administration and Anthropic continues, claims like this one will keep outpacing their own context. Rubinson’s advice for any reader scrolling past the next one is simple.
“When scrolling on social media, users should always question the source of dramatic claims before sharing,” Rubinson told Straight Arrow. “If the post you are seeing does not cite a reliable source, there’s good reason to be skeptical.”
Round out your reading
- Not red or blue: America’s politically homeless middle.
- Peter Thiel’s ‘Dialog’ network was super-secret. A data leak changed that.
- The novel legal strategy that Taylor Swift and Matthew McConaughey are using to fight AI.
- Illinois balances budget with new $200 million social media tax that tracks in-state users.
- When Trump serves up ‘Just the News,’ it comes with a side of bias.
Ella Rae Greene, Editor In Chief
