Andrew Ross Sorkin didn’t waste any time getting to the question on many people’s minds this morning when Anthropic CEO Dario Amodei appeared on-stage in Lower Manhattan with Jamie Dimon, CEO of JPMorgan and a freshly-minted Anthropic partner. “Is the freakout over AI-enabled cyberattacks warranted?” the CNBC host and veteran New York Times‘ business journalist asked.
Both men paused. The crowd laughed. Amodei looked over at Dimon, who did what powerful people do when the honest answer is uncomfortable: He began talking around it, at length, with great authority.
“Cyber is our biggest risk,” Dimon said. “It’s been our biggest risk for years.” Dimon recalled he previously warned cyber would be made worse by AI—and then came the Anthropic’s Mythos model, deemed too dangerous by AI company to release widely.
That might sound like a yes. But Dimon quickly pivoted to solutions, timelines, and reassurances about the financial sector’s preparedness. Amodei, for his part, called the risks “very real”—and then spent several minutes explaining why, if everyone responds correctly, things could actually end up better than before.
Neither man said the freakout was unwarranted. Neither said it was.
What they did reveal was alarming enough
The exchange came weeks after Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency session with Wall Street CEOs to discuss the systemic risks posed by Mythos—and as Anthropic continued to battle the Pentagon in court over the Trump administration’s decision to label the company a “supply chain risk” for refusing to allow the military to use its technology for “any lawful purpose.” It also capped a frenzied few days of Wall Street-related announcements from Anthropic, which sees the financial sector as a key customer and a key conduit for selling its models to companies elsewhere too.
Dimon and Amodei sat side-by-side in their first ever joint stage appearance, projecting calm. But the facts they disclosed told a different story about a very fluid situation.
The subtext of the exchange was, in some ways, more alarming than a simple “yes.” Amodei laid out a detailed and rapidly escalating threat timeline tied to Anthropic’s newest model. Six months ago, he said, Anthropic detected the first attempts by Chinese state-linked actors to conduct cyberattacks on U.S. tech companies using Claude, Anthropic’s AI assistant. Three months ago, a prior model found roughly 20 vulnerabilities in Firefox.
“Now, with Mythos, we found almost 300 vulnerabilities in Firefox and thousands, probably by now tens of thousands, beyond closed doors,” he said.
Amodei noted Anthropic sent someone to Congress to testify on the escalating cybersecurity risks posed by increasingly advanced AI models six months ago, and it’s been disclosing vulnerabilities as it finds them, including in the case of Firefox.
“So I think the risks are very real,” Amodei said.
The Anthropic CEO said he “would guess” the other AI labs are about one to three months behind him, while the Chinese models are “maybe six to 12 months behind.” (Although OpenAI has released a model, GPT-5.5, that in testing by the U.K. government performed almost as well as Mythos at orchestrating a complete end-to-end cyberattack.) That means we have roughly that amount of time to fix all these vulnerabilities, he said, and there are a lot of those.
“We’ve identified these tens of thousands of vulnerabilities,” he said. “The reason we haven’t announced many of them is, of course, only a small fraction have been fixed.”
That is pretty much the definition of a freakout-worthy situation.
On regulation, neither man wants the FDA model
Amodei chose not to say ‘freak out,’ but he did suggest there was a race against time to patch the vulnerabilities that Mythos and other advanced AI models are uncovering. Dimon agreed the patches themselves need to happen faster than ever.
“In the old days, you put out a patch, people had a week or two to fix it,” he said. “Now you say it’s got to be like minutes.”
Both executives were asked about a potential executive order that could create an FDA-style approval process for new AI models. Amodei was pointed in his skepticism.
“The FDA slows down medical progress a lot,” he said. “That’s a cautionary tale.”
He suggested the automotive industry—where innovation is permitted but safety standards are mandatory—would be a better model, implicitly arguing AI will need its own version of seat belt laws, crash test dummies, traffic lights, and the like.
Dimon, who has spent decades navigating financial regulation, was characteristically blunt about government overreach: “They can help us by not overburdening us when we have eight regulators asking us the same damn questions every day.”
What both men said they want is predictability—a consistent, streamlined process that applies equally to all companies, allows commercial releases to continue monthly, and targets only the most serious risks. What they have now, Amodei said, is “a bit of an ad hoc process, something that almost mimics regulation but lacks the consistency.”
The Bigger Evasion
The cybersecurity exchange was a preview of a pattern that held throughout the conversation. When asked whether AI could trigger the 20%-30% unemployment that some economists have projected, both men were less definitive than Amodei himself was in 2025, when he said half of all entry-level white-collar jobs could disappear.
“Whenever people put these things on Twitter, they only clip the part where you say the most salacious thing,” Amodei said.
Dimon was more optimistic — invoking historical precedent from agriculture to electricity to the internet — but added a caveat that undercut the reassurance: “My view of life is we don’t really know how quickly it’s going to go.”
For two of the most influential figures at the intersection of AI and global finance, that may be the most honest thing either of them said all day.
