Researchers Disclose HalluSquatting, an Attack Exploiting Hallucinated Names in 9 AI Coding Tools

On July 8, 2026, researchers disclosed a new attack called HalluSquatting that abuses the "hallucinations" of AI coding assistants to make agents fetch and run malicious code, enabling botnet assembly and more. Attackers pre-register nonexistent repository or skill names that AI models tend to invent, then plant prompt-injection instructions inside them. When an agent mistakenly fetches such a resource, a reverse shell or malware can be installed via the integrated terminal. (The Hacker News, Ars Technica)