AI Scrapes Web, Spreads Scam Phone Numbers

A new cybersecurity threat is emerging as cybercriminals exploit AI chatbots by poisoning the public web content they scrape. This tactic, dubbed "LLM phone number poisoning," involves manipulating online information to push scam call center numbers, such as fraudulent airline reservations lines. Researchers have observed campaigns actively using this method to ensure AI systems like Google's AI Overview and Perplexity's Comet recommend these scam numbers as legitimate contact details.

The technique bypasses direct attacks on AI models by compromising the data they ingest, a process akin to prompt injection. Instead of optimizing for traditional search engines, efforts are now focused on "Generative Engine Optimization" (GEO) and "Answer Engine Optimization" (AEO) to become a trusted source for AI summaries. This allows malicious content to be integrated into AI-generated answers, steering users toward scams.

This widespread contamination effect poses a systemic risk across various AI platforms. Users are advised to exercise extreme caution, always verify AI-provided contact information, and refrain from sharing sensitive data with AI assistants due to their current immaturity and security vulnerabilities.