New AI-Targeted Cloaking Attack Tricks AI Crawlers Into Citing Fake Info as Verified Facts

Cybersecurity researchers have flagged a new security issue in agentic web browsers that could expose underlying AI models to context poisoning attacks. Devised by AI security company SPLX, the attack, codenamed AI-targeted cloaking, can have bad actors set up websites that serve different content to browsers and to AI crawlers used by ChatGPT and Perplexity.

The technique is a variation of cloaking, with attackers optimising for AI crawlers via a simple user agent check that dictates what content they receive, potentially turning what AI systems regard as truth into biased or misleading outputs. According to The Hacker News, the researchers said that whatever content is served to these crawlers becomes ground truth in AI overviews, summaries, or autonomous reasoning.

The hCaptcha Threat Analysis Group noted that in many tested scenarios a lack of safeguards and technical capabilities, rather than explicit blocks, led to risky actions being executed, and it has been observed that several AI tools can perform dangerous tasks when framed as debugging exercises. The report cautions that AI crawlers can be deceived just as early search engines were, with potentially far greater downstream impacts.