Human mythology’s cybersecurity concerns: What smart people say

Anthropic’s announcement this week about a powerful new AI model sparked a wave of warnings and dire predictions, but not everyone is buying the hype.

Anthropic announced Tuesday that it will not release its next-generation AI model, Mythos, due to cybersecurity concerns. The company says Mythos is so powerful that it could be used by non-experts to exploit major operating system vulnerabilities.

Instead of a wide release, Anthropic said it will make the Claude Mythos Preview available to 11 external organizations, including Google, Microsoft, Amazon Web Services, JPMorganChase and Nvidia, as part of “Project Glasswing.”

Mr. Anthropic’s claims about Mythos’ capabilities quickly raised concerns, as did his meetings with Federal Reserve Chairman Jerome Powell, Treasury Secretary Scott Bessent and the heads of major U.S. banks.

While some AI critics have warned about the cybersecurity implications, others have questioned the significance of Anthropic’s announcement, saying it doesn’t seem like Mythos is leaps and bounds ahead of other models and is more likely a matter of good PR.

Should Mythos make security professionals tremble? Is Anthropic just a master at marketing its model? As the internet debates the latest AI developments, here’s what smart people are saying.

gary marcus

AI researcher and author Gary Marcus said Anthropic’s announcement about Mythos was “overblown.”

“To some extent, I feel like we were being played,” Marcus wrote on Substack. “This demonstration was definitely a proof of concept that we need to have regulatory and technical institutions in place, but it was not the immediate threat that the media and the public were led to believe.”

Marcus said that from what he’s seen so far, the model appears to be “incrementally better” than previous models rather than “groundbreaking.”

Yann LeCun

Yann LeCun, founder of AMI Labs and former chief AI scientist at Meta, also poured cold water on the Mythos hype.

“Mythological drama = BS from self-delusion,” he wrote in X’s post.

He was responding to a post from AI security company Aisle, which said it tested a smaller, cheaper model for the same vulnerabilities highlighted in Anthropic’s Mythos announcement and found it could perform much of the same analysis.

jake moore

Jake Moore, global cybersecurity specialist at ESET, previously told Business Insider that while Anthropic’s announcement contained marketing language, “fundamentally this model is incredibly impressive and will only get better over time.”

“Anthropic has built a reputation as a ‘safety first’ AI company, so announcements like this serve two purposes: to take real notice and to demonstrate the company’s commitment to safety,” Moore said.

Dave Kasten

Dave Kasten, director of policy at Palisade Research, said he believes other AI models are unlikely to come close to Mythos.

He said in an interview with CNBC on Thursday that he expects “Anthropic to have a little bit of a lead, but not a huge lead, and there’s not necessarily a permanent moat here.”

He noted a recent report from Axios that said OpenAI also has models with advanced cybersecurity features and plans to release them only to a small group rather than the general public.

Kasten also said that while Gemini is probably not far behind, the fact that Google is partnering with Mythos suggests that Anthropic will likely have an advantage in this particular model for at least a few months.

david sax

David Sachs, a tech investor and former White House AI czar, said Anthropic’s claims about Mythos are important but should be taken with a grain of salt.

“The world has no choice but to take the cyberthreats associated with Mythos seriously, but we cannot ignore Anthropic’s history of fear tactics,” Sachs said in the X post, sharing several past examples of Anthropic issuing alarming warnings and clarifications regarding its AI models.

TJ Marlin

TJ Marlin, CEO of Guardrail Technologies, who previously worked in EY’s global forensic technology practice, said the meeting between federal regulators and Wall Street was meant to ensure that banks wouldn’t flip-flop and say “I didn’t know” if a major security breach occurred.

“Every CEO in the room who does not have a documented board-level response is operating in the most legally vulnerable position,” Merlin wrote on LinkedIn.

Pavlos Holman

Deep Future venture capitalist Pavlos Holman said cybersecurity defenders, those trying to defend against digital attacks, stand to benefit more from advances in AI than those who carry out the attacks.

“Right now, everyone is losing their mind over AI attacks,” Holman said in a LinkedIn post about Mythos on April 1, ahead of Anthropic’s announcement this week. “What they lack is that the defenders have the same AI, often better AI, and much more computing power.”

Holman said cybersecurity defenders will have access to the same models, as well as more resources such as source code.

“This is still an escalating war, but now the defenders have the advantage,” he wrote. “Security is about to get better, not worse.”

Ben Seri

“We have entered the Manhattan Project moment of cybersecurity,” Ben Seri, co-founder of cybersecurity startup Zaffran Security, said in a post.

Seri said that while the cybersecurity threat is real and immediate, the potential for protection is real but will take time to materialize. The real challenge, he said, is enabling cybersecurity defenders to work at scale and quickly.

“AI finds vulnerabilities faster and fixes them faster. But the bottleneck is never just discovery and remediation. It’s the ability to deploy fixes into production safely, quickly, and at scale.” “Ensuring that we adapt to rapid changes in production is the most important change technology and security leaders need to address to meet this moment.”

Alice Tekotski contributed reporting.