Yann LeCun, former chief AI scientist at Meta and one of the three researchers credited with pioneering modern deep learning, dismissed the panic surrounding Anthropic’s Claude Mythos Preview as far-fetched drama. “Mythic drama = self-delusional BS,” LeCun wrote on X, responding to a post by AI security company Isle that claims it can reproduce much of the same vulnerability analysis that Anthropic released this week in a smaller, cheaper model.LeCun’s comments come days after Anthropic announced that Mythos Preview had discovered thousands of high-severity vulnerabilities across all major operating systems and web browsers, a claim serious enough to prompt Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent to call an emergency meeting with the CEOs of the nation’s largest banks. The model, which Anthropic declines to disclose publicly, is being shared only with a select group of companies, including Apple, Google, and Microsoft, under a $100 million initiative called Project Glasswing.
LeCun is not alone in calling this hype.
LeCun’s skepticism is not isolated. Gary Marcus, an AI researcher and frequent industry critic, wrote on Substack that the Mythos threat was “overblown” and that “to some extent we feel we were fooled.” Marcus argued that the model is not groundbreaking and appears to be “incrementally better” than previous models. Former White House AI czar David Sachs acknowledged the cybersecurity risks, but added, “It’s hard to ignore that Anthropic has a history of blackmail tactics.”Palisade Research’s Dave Kasten told CNBC that Anthropic is “a little bit ahead, but not overwhelmingly ahead” and doesn’t have a permanent moat, noting that OpenAI has a model with similar capabilities that it reportedly plans to limit.Timing didn’t help with Anthropic. The company’s expected annual revenue this year has tripled to more than $30 billion, and both Anthropic and OpenAI are reportedly considering IPOs. A model too dangerous to sell, available only to the world’s biggest companies, whose safety decisions can be read as easily as sales pitches.
But companies that actually use Mythos tell a different story.
Not everyone agrees with LeCun. Organizations that have been implementing Mythos Preview for weeks (Cisco, CrowdStrike, Palo Alto Networks) are treating it as a true turning point. “What used to take months can now be done in minutes with AI,” said Elia Zaitsev, CrowdStrike’s chief technology officer. Cisco’s chief security officer called this a breaking point “of no return.” These are cybersecurity companies with their reputations on the line, not Anthropic’s marketing arm.
