Mythos Preview: AI Powers “Vulnpocalypse,” Turning Hackers Into Superweapon Users as

Anthropic has limited access to Mythos Preview to roughly 50 firms under “Project Glasswing,” the company said, in an effort to protect critical software, NPR reported. The model can locate high‑severity flaws in every major operating system and web browser and also generate exploitation tactics, making it a potent tool for both defenders and attackers.

The decision to withhold the model sparked a rapid response from U.S. officials. Treasury Secretary Scott Bessent called a meeting with leading banks this week to discuss the “rapid developments taking place in AI,” a Treasury spokesperson confirmed, citing concerns that the technology could be weaponized against financial infrastructure, NBC News said.

Security researchers warn that the same capabilities could trigger a “Vulnpocalypse,” where hackers use AI to accelerate attacks on hospitals, factories and other critical services. Kevin Collier of NBC News noted that experts see the scenario moving from theory to reality after Anthropic’s announcement.

Anthropic’s CEO Dario Amodei highlighted the dual‑use nature of the technology during the World Economic Forum in Davos, emphasizing that the model’s ability to suggest exploit paths could dramatically increase malicious success rates, according to the NBC News interview.

For now, Anthropic is sharing Mythos Preview only with select tech giants and partners, hoping the limited rollout will help shore up defenses before the model inevitably spreads, NPR wrote. The company’s cautious approach underscores the growing tension between AI‑driven security innovation and the risk of creating a new class of cyber superweapon.