The leak of Claude Mythos, Anthropic’s AI model “more advanced than any other in cyber capabilities”, has shaken Wall Street and reopened the most disturbing question of the digital age: who really controls artificial intelligence?
There is news that shakes the markets, and news that shakes something deeper: our idea of where we are headed. In what appears to be a significant security incident, Anthropic has unintentionally revealed details about an AI model in development, called “Claude Mythos”through its company website. This was not a hacker attack planned by a hostile state. It was something much more banal, and in a way much more disturbing: a human error.
A configuration error in Anthropic’s content management system accidentally exposed a draft blog describing the template, available in an unsecured, publicly searchable data repository, stating that the new template «poses unprecedented cybersecurity risks».
Anthropic has not officially confirmed all the details, but has acknowledged the existence of an information leak, launching an internal investigation. The damage, however, was already done.
What is Claude Mythos
The leaked document describes Claude Mythos under the product name “Capybara”: a new model level that would sit: bigger and smarter than current Opus modelswhich until now were the most powerful. Compared to the previous best model, Claude Opus 4.6, the new system would score significantly higher on tests of software coding, academic reasoning and cybersecurity.
According to Techzine who had access to Antrhopic’s draft document, the model is described as “currently far ahead of any other AI model in cyber capabilities” and warns that “heralds an imminent wave of models capable of exploiting vulnerabilities in ways that far exceed the efforts of the defenders.” In the days that followed, the disaster worsened: Anthropic accidentally leaked Claude Code’s source code as well, exposing around 500,000 lines of code in around 1,900 files.
Panic on the stock market
The markets don’t wait for certainties. They react to perceptions, to fears. Cybersecurity stocks fell sharply following the release of the leaked draft. Companies like CrowdStrike, Palo Alto Networks And Okay have seen significant declines. Investor logic anticipates a brutal and dramatic sentiment: If an AI model can identify and exploit vulnerabilities better than any human team, are traditional security companies at risk of failing spectacularly?
Analysts at DA Davidson said they believe highly unlikely that AI labs can replace cybersecurity vendors, noting that most Chief Information Security Officers remain highly skeptical. Yet no analyst can ignore the trajectory: The speed with which models increase in power, and their increasing accessibility to anyone — including malicious actors, are a real and far from trivial threat.
David versus Goliath?
The point is not just what Mythos can do, but who might use it and with what intentions. On the other hand, it is clear that sophisticated artificial intelligence tools can significantly reduce the time to develop sophisticated cyber attacks and similarly limit risks with equally effective tools.
AI doesn’t level the playing field, it tilts it. Defenders must protect every single vulnerable point. Attackers only need to find one. As companies roll out increasingly advanced models, AI agents learn to act and reason without human intervention, allowing hackers to conduct multiple attack campaigns at once that are much more difficult to defend against.
We live in a civilization radically dependent on digital infrastructure. Hospitals, electricity grids, banking systems, food chains: everything is connected, everything is vulnerable. The Claude Mythos affair is a window onto the precipice. Not because Anthropic is irresponsible, but because it shows that we are moving fast, too fast, and the risk is that of lose control of the vehicle.
International AI governance systems are years behind. Existing regulations were designed for much less capable AI. And geopolitical competition creates strong incentives to accelerate further rather than slow down. The apocalyptic scenario is not science fiction: it’s a real possibility, one that gets closer with each new model released.
