In controlled evaluations, the model demonstrated the ability to identify vulnerabilities and execute multi-step attack strategies that would typically take human experts many hours. In one simulation, it completed a complex 32-step corporate network attack in multiple attempts, showcasing a level of autonomy that has raised concerns among security professionals.

The implications are significant. While the model is not publicly released, its capabilities have already triggered discussions among policymakers and financial leaders about potential cyber risks. It highlights a growing reality: AI is no longer just a defensive tool—it can also be a powerful weapon if misused.