Cybersecurity & Intelligence — 2026-03-28
Anthropic Data Leak Exposes Unreleased Claude Mythos Model with Unprecedented Cybersecurity Risks
Anthropic accidentally exposed ~3,000 unpublished assets through an unsecured content management system, revealing details of an unreleased AI model called Claude Mythos that the company itself describes as posing "unprecedented cybersecurity risks." The model represents a "step change" in AI capabilities. A US judge separately blocked the Pentagon's attempt to ban Anthropic from government work. Cybersecurity stocks sold off on the news, with the market implications compounding broader war-driven declines.
Analysis
Prior cycle reported the $45 billion cybersecurity stock selloff triggered by the initial leak. The new detail this cycle, a federal judge blocking the Pentagon's attempt to ban Anthropic from government work, means the Department of Defense will continue deploying AI systems from a company that just demonstrated a significant operational security failure. A model flagged for cybersecurity risks was exposed through a basic CMS misconfiguration, revealing systemic gaps in AI governance.
Prior cycle reported the $45 billion cybersecurity stock selloff triggered by the initial leak. The new detail this cycle, a federal judge blocking the Pentagon's attempt to ban Anthropic from government work, means the Department of Defense will continue deploying AI systems from a company that just demonstrated a significant operational security failure. A model flagged for cybersecurity risks was exposed through a basic CMS misconfiguration, revealing systemic gaps in AI governance.