Researchers have identified a new technique dubbed "AI as a C2 proxy," where AI assistants like Microsoft Copilot and xAI Grok are exploited as stealthy command-and-control (C2) relays. By leveraging these tools' web-browsing and URL-fetching capabilities, attackers can create a bidirectional communication channel that blends into legitimate enterprise traffic. This method bypasses traditional defenses like API key revocation because it utilizes anonymous access to retrieve attacker-controlled instructions and exfiltrate data.
Beyond simple communication, these AI models can be used to dynamically generate reconnaissance workflows and adapt malware behavior at runtime. This evolution moves toward AI-driven implants that can automate triage and operational choices in real-time. Related research also highlights how LLM APIs can be used to generate malicious JavaScript dynamically, effectively bypassing network security controls through client-side assembly.
Top comments (0)