New hack exploits AI hallucinations to trick agents into running malicious code — 'HalluSquatting' attack exploits a fundamental weakness in every available model
Attackers can exploit how AI bots hallucinate software URLs to create massive botnets.
The vulnerability is endemic to every model.