1-minute read
What a defender needs to know
Push Security documented a malvertising campaign that sent people from paid search results to shared pages on ChatGPT and Claude. One ChatGPT page used rendered HTML and CSS to imitate a service disruption notice. Its download button led to an attacker-controlled clone of the desktop-app page.
The real platform hostname gave the first page credibility, but it said nothing about who created the shared content or where the next click would go. Push also found that the external site showed a fake download page to interactive browsers and a benign company page to URLScan. Shared Claude conversations used a related lure that told macOS users to run a malware-delivery command.
Defenders should examine the browser route as a sequence. Look for shared AI pages reached from search results, cross-domain clicks that lead to installers and terminal commands copied from public conversations. Review recent AI-branded downloads, preserve file and process evidence, and give employees an easy way to verify official installation paths.
Public AI sharing pages should be treated like any other user-generated content. The platform can be legitimate while the page and its links remain hostile.