Shared ChatGPT Link Flaw Could Expose Private Chats

Researchers identified a chain of three design flaws in ChatGPT’s shared link feature that allowed code execution within an authenticated user session. The attack relied on a crafted shared link, demonstrating that a user clicking a seemingly passive collaboration element could trigger unintended actions under their own signed-in identity.

The finding underscores that collaboration features designed for static sharing require the same isolation and origin checks as active application components. The flaws were chained together to move from a shared resource to code execution inside the target’s session context.