A single security lapse at Anthropic’s headquarters last week revealed a monster living in the code: 'Mythos,' an AI model capable of autonomously uncovering 20-year-old vulnerabilities in seconds. Rather than releasing this tool to the public to fortify the digital commons, Anthropic has locked it behind 'Project Glasswing.' This elite perimeter grants access only to 40 of the world’s most powerful entities, including JPMorgan Chase and the Pentagon. This is 'Cognitive Enclosure'—the privatization of the very tools needed to defend against the future.

By restricting Mythos, Anthropic has established a new kind of digital feudalism. The entities inside the circle gain a god-eye view of every crack in the world’s software, while the public remains vulnerable to the same exploits. A veteran security researcher, speaking on condition of anonymity, described the move as "seizing the fire and selling the smoke." The common user is left to rely on 'less capable' models like Claude 4.7, while the state and corporate elite wield the master key.

The source code leak that first exposed Mythos has already triggered a global auditing supercycle, but it is an audit for the few. As AI moves from a tool to an autonomous actor, the ability to find and fix vulnerabilities is the ultimate form of sovereignty. By gating that ability, tech giants are ensuring that the digital perimeter is a private asset, not a public utility. The software world is being divided into those who can see the bugs and those who must simply wait to be hit by them.