AI · 108 days ago
The break is not the fake link. It is that the lure can sit inside a real chatgpt.com page, so users see a trusted AI interface and let their guard down. That also weakens browser reputation checks and desktop-only download controls, because the malicious content is delivered from the platform the user already trusts.
Researchers disclosed ChatGPhish, which abuses ChatGPT’s Markdown rendering so attacker-controlled links, images, and security-style alerts appear inside the assistant UI after a page is summarized. BleepingComputer also reported an active LLMShare campaign using shared chatgpt.com links to show fake OpenAI outage pages that push malware downloads from a legitimate OpenAI domain.
The forward risk is platform trust itself. Any workflow that renders untrusted content inside a SaaS page can turn the vendor domain into the phishing surface, not just the destination URL.
2 sources covering this story
ChatGPhish Vulnerability Turns ChatGPT Web Summaries Into a Phishing Surface
ChatGPhish exploits ChatGPT Markdown rendering to deliver phishing content from summarized web pages, increasing AI attack surfaces.
ChatGPT share links abused to host fake outage pages to deliver malware
Threat actors are abusing ChatGPT's content-sharing feature to display fake OpenAI outage pages that direct users to download malware disguised as the ChatGPT desktop application.
Part of the PlainSec briefing for 2026-05-30