Palo Alto Networks flags HalluSquatting: hackers register fake AI-made web addresses

Palo Alto Networks’ Unit 42 is warning about a chatbot security problem that’s gaining ground: HalluSquatting. AI systems are making up domains, download links, and package names, and attackers can move first by registering them.

Avast Free Antivirus Download

Unit 42 researchers say the play is pretty simple. When a model invents a website or software package that sounds plausible, criminals can turn that fake asset into phishing pages, credential theft, malware delivery, or supply-chain attacks. The risk gets higher when people copy and paste generated links, or when automated workflows trust them without checking.

In its analysis, Palo Alto Networks’ Unit 42 looked at 2.1 million URLs generated by two large language models across 913 global brands. The team found more than 13,000 confirmed malicious URLs, plus roughly 250,000 hallucinated domains that were still unregistered.

What makes the problem hard in practice, Unit 42 says, is that different models often hallucinate the same names. One malicious registration can pull in traffic from developer tools and customer-facing chatbots across a lot of different places. In one documented case the team cited, a coding assistant even helped put together a phishing kit on a predicted phantom domain.

If you use AI tools often, this isn’t something to shrug off. Public forecasts now put machine-learning-driven cyberattacks at more than 28 million in 2025, and public reports say 35% of botnet operations used ML to evade detection, even if those same reports haven’t yet shown HalluSquatting by itself producing massive botnets.

The takeaway from Palo Alto Networks’ Unit 42 is straightforward: verify generated domains, packages, remediation steps, and any link that merely looks convincing.