Researchers at Check Point Software Technologies demonstrated that modern AI assistants with browsing and URL-fetching features misused as covert command-and-control relays. Services such as Microsoft Copilot and Grok normally blend into trusted enterprise traffic, making them attractive intermediaries for malware communications. Instead of contacting attacker infrastructure directly, implants can route requests through these AI web interfaces, which fetch attacker-controlled content and return results as benign summaries, masking malicious traffic within legitimate AI usage, complicating detection, blocking. Firstly, attackers deploy a lightweight implant that embeds or downloads the WebView2 runtime to emulate a legitimate browser session. It enumerates host data including usernames, domain membership, installed software, running processes, and security controls; crafts prompts instructing the AI to retrieve a remote webpage in attacker’s control. Sensitive reconnaissance details are encrypted or encoded within query parameters to bypass filtering. Because the AI service handles the outbound request, network monitoring tools perceive only routine connections to trusted AI domains rather than suspicious communication. In the command phase, the attacker’s site returns structured content embedding instructions within text/HTML patterns. The AI assistant summarizes or extracts this content, presents it as a normal response. The implant parses the output, decodes commands, executes tasks like downloading payloads, running PowerShell, staging data exfiltration. This loop enables bidirectional exchange: results are encoded, sent back through prompts, and relayed to the attacker. The model can also assist decision-making by prioritizing targets, identifying valuable files, adapting tactics dynamically, creating context-aware, AI-driven operations. Defensively, organizations should treat AI domains as high-risk egress paths and apply zero-trust monitoring. Enforce authentication, restrict anonymous browsing features, inspect automated WebView usage. Implement behavioral detection, network segmentation, strict outbound controls, and encrypted backups. Providers like Microsoft must harden fetch capabilities and add auditing, while security teams combine patching, logging, and threat hunting to reduce abuse and impact.
Security researchers have identified a new variant of the SparkCat malware circulating on both the Apple App Store and Google Play Store, more than a year after its initial discove...
The European Union’s cybersecurity agency, CERT-EU, has attributed a significant cloud breach involving the European Commission to the TeamPCP threat group. The attack targeted t...
Microsoft has rolled out automatic updates for unmanaged Windows 11 24H2 Home and Pro edition devices to Windows 11 25H2, also known as the Windows 11 2025 Update. This transition ...