AI Model Flags Vulnerabilities in Classified U.S. Systems During Security Test
Published on: June 25, 2026
In a recent security evaluation, Anthropic’s advanced AI model, Mythos, successfully surfaced vulnerabilities within classified U.S. government computer systems in a matter of hours. This occurred as part of a collaborative initiative known as "Project Glasswing," which involves partnerships between Anthropic and U.S. intelligence agencies to assess potential national security and public safety risks associated with cutting‑edge AI systems.
A U.S. official, speaking anonymously, confirmed that while Mythos identified the vulnerabilities swiftly, it did not necessarily exploit them during the testing phase. The speed and precision with which the model operated has raised both intrigue and concern among cybersecurity and government observers.
Senator Mark Warner of Virginia highlighted the severity of the findings during a Senate hearing on June 11, quoting the head of the National Security Agency: “This tool broke into almost all of our classified systems, not in weeks but in hours.” The NSA and Anthropic declined to comment further on the matter.
This development comes amid growing tension between Anthropic and the current administration. The administration recently issued a directive restricting foreign nationals' access to Anthropic’s advanced models, Fable 5 and Mythos 5. Anthropic temporarily disabled access to these models to comply, maintaining that the government’s concerns were overstated relative to the security risk identified.
The episode underscores a broader debate regarding AI’s dual-use nature: while models like Mythos can significantly enhance cyber-defence by pinpointing software flaws, they may also accelerate offensive cyber capabilities if misused. A letter signed by over 100 experts from companies including Adobe and Nvidia urged reconsideration of limiting access to such tools, warning that sidelining them could undermine United States’ ability to defend against adversaries.
In response, the situation highlights the urgent need for thoughtful governance frameworks around frontier AI models—balancing innovation, national security, and ethical use. As AI becomes increasingly capable of rapid and sophisticated system analysis, policymakers face mounting pressure to establish oversight mechanisms that keep pace with technological advancement.
No comments yet.