AI code assistants face new HalluSquatting botnet threat discovered by researchers

Security researchers from Intuit, Technion and Tel Aviv University have identified a novel attack method called HalluSquatting. The technique exploits the hallucination tendency of large language models used in AI coding assistants, manipulating them to fetch malicious resources that can turn users' computers or smartphones into botnet nodes.

Tests show that popular tools such as GitHub Copilot, Cursor, Gemini CLI, Windsurf, Cline and several others are vulnerable. By injecting harmful commands into seemingly legitimate AI‑suggested sources, attackers can achieve remote code execution, gain full control of the device, and even launch large‑scale ransomware campaigns.

Mitigation recommendations include restricting external data fetching by AI tools, enforcing stricter naming conventions for resources, and fostering coordinated security efforts among platform providers.