ChatGPT's deep research tool leaked users' private Gmail data

Technology Sep 19, 2025

A big security slip in OpenAI's ChatGPT was detailed in a report published by Radware on September 19, 2025.
Researchers demonstrated a trick called "Shadow Leak"—basically sneaky instructions hidden in emails—to get private Gmail info through the Deep Research tool.
Since the AI was acting on its own, normal security checks didn't catch it.

Other apps like Google Drive, Outlook also at risk

Radware pointed out that other apps connected to Deep Research, like Google Drive and Outlook, could have been at risk too.
OpenAI quickly fixed the problem, but this whole episode is a reminder for tech companies to stay sharp with their security and respond fast when issues pop up.