Executive Summary
Unit 42 has identified a novel supply chain vector termed 'phantom squatting,' where threat actors register domains that Large Language Models (LLMs) consistently hallucinate as legitimate endpoints. These 'phantom domains' provide a structural advantage for attackers because they possess zero reputation at the time of registration, bypassing legacy blocklists and reputation-based security filters.
Research reveals that LLMs, including major enterprise and frontier models, generated over 250,000 unique unregistered domains during testing. A significant real-world case involved the Montana Empire phishing kit, which targeted customers of a national postal service marketplace using a hallucinated domain predicted by researchers 23 days prior to its registration. This represents a closed-loop threat where AI is used both to generate the delivery vector and to assist in developing the malicious tooling.