OpenAI CEO Sam Altman took aim at Anthropic’s new cybersecurity model, Mythos, during a recent Core Memory podcast. Anthropic rolled out Mythos this month to a handful of enterprise clients, arguing it’s so powerful that public release would let cybercriminals weaponize it. Altman dismissed that pitch as fear-based marketing designed to keep advanced AI tools exclusive. He compared Anthropic’s tactic to selling a bomb shelter after claiming you’ve built a bomb. Altman said some players in the industry have long wanted to restrict AI access to a privileged few. By stoking doomsday scenarios, companies can justify high-priced, limited releases. He pointed out that this approach isn’t unique to Anthropic—plenty of AI vendors, himself included, have leaned on hyperbole about AI’s risks to drive demand. Anthropic’s claim hinges on Mythos’s supposed ability to plan and execute sophisticated cyberattacks. But critics inside and outside the industry argue the model’s threat is overstated. They note that most real-world hacking still relies on human actors and simpler tools. Even Anthropic users who’ve tested Mythos haven’t reported breakthroughs that couldn’t be matched by existing software or hacking frameworks. By framing Mythos as a world-ending weapon, Anthropic draws attention and fees from big corporations worried about cybersecurity. Altman’s counterargument is straightforward: if fear is the hook, only a small elite gets to deploy the tech. He wants AI development to be more open, even if that means acknowledging and managing risks rather than exaggerating them.