Check Point Research has identified a new evolution in cyber threats where AI assistants like Grok and Microsoft Copilot are being exploited as covert Command and Control (C2) relays. Unlike traditional malware that relies on static code, AI-Driven (AID) malware utilizes the web-browsing capabilities of AI agents to tunnel data through legitimate corporate traffic. By using browser emulation techniques like WebView2, attackers can establish a bidirectional communication channel that avoids the need for API keys or registered accounts, effectively bypassing traditional security controls like account suspension or key revocation.
Beyond simple communication, this research highlights the shift toward 'AIOps-style' C2, where AI models act as remote decision engines for the malware. This allows the implant to perform context-aware actions such as environment validation to evade sandboxes and prioritized data exfiltration based on the perceived value of specific files. As AI becomes more integrated into enterprise environments, defenders must treat AI service domains as high-value egress points and update detection pipelines to account for these adaptive, model-driven tactics.
Top comments (0)