An unexpected leak has put Anthropic's most advanced artificial intelligence at the center of debate. The model, known as Claude Mythos, has raised concerns across both the technology sector and financial markets because of its potential offensive cybersecurity capabilities.
## A leak that should never have happened
According to multiple reports, thousands of internal documents were exposed in a public repository due to a configuration error in Anthropic's content management system. The company attributed the incident to "human error," although the scale of the leak has raised doubts across the community.
Security researchers were able to access more than 3,000 files, revealing technical details about the company's most advanced model.
## Claude Mythos and the jump to the "Capybara" category
The leaked model, Claude Mythos, belongs to a new internal category called Capybara, designed to significantly outperform earlier versions such as Claude Opus 4.6.
The documents describe major improvements in advanced reasoning, software development, and vulnerability analysis.
## Internal risks: the company itself warns of the danger
The internal documents themselves acknowledge serious risks, including automated vulnerability discovery, the ability to exploit systems autonomously, and the potential orchestration of complex cyberattacks.
## Immediate market impact
Investor reaction was immediate. Companies such as CrowdStrike, Palo Alto Networks, and Zscaler posted notable stock declines.
The market fears that models like this could shift the balance between attackers and defenders.
## An uncertain future
Anthropic has confirmed that it is developing a next-generation model, but it has not announced a release date. The company says it is proceeding cautiously because of the implications of this technology.
## Conclusion
The Claude Mythos leak points to a possible paradigm shift in cybersecurity. As artificial intelligence advances, the line between defense and offense becomes increasingly blurred.