- Anthropic accidentally exposed internal documents revealing Claude Mythos, a new AI model described as a “step change” in performance over its predecessor, Opus 4.6.
- The leaked files warn that the model poses “unprecedented cybersecurity risks” due to its advanced ability to identify and exploit software vulnerabilities at scale.
- The news triggered a significant market reaction, with cybersecurity stocks like CrowdStrike and Palo Alto Networks falling as much as 6%.
- Bitcoin and broader crypto markets saw a brief tumble back to the $66,000 range following the disclosure of the security lapse.
Anthropic, the AI safety-focused firm backed by multibillion-dollar investments, inadvertently leaked sensitive internal documents on Thursday evening, revealing the existence of its most powerful model to date: Claude Mythos. The exposure, attributed to a misconfiguration in the company’s content management system (CMS), left nearly 3,000 unpublished assets—including draft blog posts and risk assessments—accessible in a public data lake. The leaked materials describe Mythos as a breakthrough in software programming, academic reasoning, and cybersecurity, sitting within a new high-end performance tier dubbed “Capybara.”
The market’s immediate concern centers on the model’s reported autonomous exploit capabilities. According to the leaked drafts, Anthropic’s own internal testing suggests Claude Mythos “presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.” This admission from a company that prides itself on “AI alignment” and safety has rattled both traditional and decentralized finance sectors. Investors fear that such a tool could be weaponized to target smart contract vulnerabilities or financial infrastructure with a speed that human-led security teams cannot match.
Wall Street responded with a sharp “AI scare trade.” Major cybersecurity firms, including Palo Alto Networks (PANW), CrowdStrike (CRWD), and Zscaler (ZS), saw their shares slide between 4% and 7% on Friday. Analysts suggest the sell-off reflects a growing anxiety that advanced AI agents might render current deterministic security models obsolete or dramatically lower the barrier for sophisticated cyberattacks against enterprise and crypto-native platforms.
In response to the leak, an Anthropic spokesperson confirmed the model’s existence, stating: “We are developing a general-purpose model that has made substantial progress in reasoning, coding, and cybersecurity. Given the model’s powerful capabilities, we are adopting a cautious approach to its release.” The company noted that Mythos is currently undergoing limited trials with early access customers, primarily within the defensive security community, to bolster codebases before a wider rollout.
The incident also highlights a persistent irony in the frontier AI sector: the very companies developing super-intelligent security systems remain vulnerable to basic operational security (OpSec) failures. This leak follows a recent $30 billion funding round for Anthropic, and it is expected to intensify regulatory calls for mandatory security audits of AI developers handling models with national security implications.
Disclaimer: This article is for informational purposes only and does not constitute advice of any kind. Readers should conduct their own research before making any decisions.
