Do you trust the answers you receive from AI? Studies show that 92% of users accept AI-generated answers as established fact without verifying them. Now, Israeli cybersecurity researchers - Dan Lasker, Ariel Simon, and Naor Haziz, all veterans of elite Israeli army intelligence units, have uncovered how scammers exploit this consumer blind spot to profit at the expense of unsuspecting consumers.
According to a study published last week in the US, hacker groups are running automated campaigns that contaminate the databases relied upon by search engines and chatbots such as ChatGPT, Gemini, and Google AI Overview. The attackers go beyond mere spam, employing advanced marketing techniques known as Generative Engine Optimization (GEO) to bypass technological filters, they embed phone numbers containing emojis and special Unicode characters - elements that AI engines still interpret as reliable information.
The result: innocent searches for flight cancellations or a branch's operating hours lead the customer to a phone number that appears completely legitimate. The customer calls and is connected to a professional-sounding call center that offers to assist with legitimate tasks but charges prices and fees far above the official rates.
Researchers have developed a specialized system that has so far detected active attacks targeting 374 well-known brands. These attacks primarily target major global airlines (Delta Airlines, Lufthansa, United Airlines, Emirates, Qatar Airways), travel platforms (Airbnb and TripAdvisor), and banks (Chase, Citi, Wells Fargo). Thousands of new malicious posts appear daily on government and academic (.edu) websites as well as social media networks. Even when these posts are removed, cached copies (such as those on archive.org) continue to feed AI engines with "contaminated" data, rendering any individual takedown effort futile.
No man's land
The full report reveals a global "security vacuum" in which corporations and tech giants are aware of the phenomenon yet shrugged off responsibility, leaving consumers exposed. When researchers contacted the security teams at some of the affected companies, most claimed that since their internal systems had not been breached, addressing the issue was neither within their capacity nor their responsibility.
At the same time, the researchers proactively reported their findings to Google and OpenAI but met with refusals that highlight the inadequacy of existing security models. Google, for instance, responded that reports regarding AI-generated misleading information were classified as "out of scope" for its Vulnerability Reward Program (VRP). OpenAI closed the report on the grounds that the findings were "not reproducible" - arguing that language model responses vary from one query to the next - despite the fact that the researchers had provided concrete, documented evidence of active scams. Dan Lasker, cyber vulnerabilities researcher and partner in the study said, "People have learned to be wary of suspicious links in emails and text messages, but when a phone number is presented as fact directly by an AI engine - and corroborated by multiple sources - it is perceived as entirely trustworthy. Companies themselves invest billions in protecting their servers from breaches, yet this phenomenon demonstrates that such measures are no longer sufficient. As long as tech giants refuse to acknowledge AI-generated disinformation as a genuine security vulnerability, consumers remain completely unprotected. This is the systemic gap we are working to close today."
Published by Globes, Israel business news - en.globes.co.il - on September 28, 2026.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2026.