Hackers exploit ChatGPT share links to deliver malware via legitimate domains
Threat actors abused ChatGPT and Claude sharing features to host phishing pages on legitimate OpenAI and Anthropic domains, using Google ads to route victims to fake service disruption notices that redirect to malware downloads. The attack exploited a design feature that renders HTML code in shared pages, bypassing corporate firewalls and URL-reputation checks because the malicious content appeared on trusted domains.
What happened
Threat actors purchased Google search ads targeting queries such as 'ChatGPT,' 'ChatGPT desktop app,' and 'ChatGPT download' to route victims to malicious share links hosted on chatgpt.com and Claude sharing pages.
The attackers exploited ChatGPT's code-rendering feature in the sharing system to build pixel-perfect fake service disruption pages claiming temporary unavailability due to high traffic and directing users to download the desktop app.
The malicious pages bypassed corporate web filters and URL-reputation checks because they were hosted on legitimate OpenAI and Anthropic domains. A second phishing page used conditional rendering to show benign content to automated scanners while displaying a fake download page to real users.
Reported impact
Push Security reported the campaign, named LLMShare, was generating active detections and targeting both ChatGPT and Claude users as of May 29, 2026.
Victims clicking the malicious ads were directed through fake outage notices to phishing pages designed to install infostealer malware or InstallFix malware variants.
Index analysis
The attack exploited a design feature rather than a software vulnerability in the sharing systems, meaning no patch could address it without changing how the products work.
As of June 1, 2026, neither OpenAI nor Anthropic had issued a public statement addressing the abuse of their platforms' sharing features.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.