Find notable cyber news and cases, enriched with sources, timelines, and signals.
Home Watchlists Beta Browse News Feed Stats About Contact

ChatGPT planted-instruction cross-account data exfiltration security flaw

Vulnerability
First reported
Last updated
Happening score
H score 25
1 unique sources, 1 articles

Summary

Hide ▲

ChatGPT was shown to accept a planted instruction that could trigger hidden tool use and cross-account data exfiltration from connected apps, including Gmail. In the proof of concept, the model kept answering normally while covertly reading session data and relaying it through a hidden channel to another ChatGPT account. The weakness depended on the session's existing permissions and connected services, so impact scaled with what the conversation could already access. OpenAI said the internal service behind the channel was taken offline, and there is no user update to install.

Related Happenings

OpenAI ChatGPT Work partial outage causing elevated errors

Service Disruption
H score0 First: 31.08.2026 19:50 Last: 31.08.2026 19:50 Sources 1

About this happening: OpenAI's ChatGPT Work is in a partial outage, leaving some users unable to start or continue tasks and driving elevated latency and errors. The issue began at 11:04...

Open Happening

OpenAI Anthropic and Google reasoning APIs cross-session replay security flaw

Vulnerability
H score36 First: 12.08.2026 14:47 Last: 12.08.2026 14:47 Sources 1

About this happening: A newly disclosed cross-session replay flaw in OpenAI, Anthropic, and Google reasoning APIs exposed hidden reasoning and secrets from session logs, including API keys...

Open Happening

A coordinated network of ChatGPT accounts likely originating from Southeast Asia campaign activity escalates

Campaign
H score29 First: 05.08.2026 21:33 Last: 05.08.2026 21:33 Sources 1

About this happening: OpenAI disrupted a Poipet-linked scam campaign that used ChatGPT to scale fraud across investment, romance, gambling, and law-enforcement impersonation schemes. The di...

Open Happening

ChatGPT Workspace Agents CSRF AgentForger security flaw

Vulnerability
H score40 First: 24.07.2026 14:53 Last: 24.07.2026 14:53 Sources 1

About this happening: OpenAI's ChatGPT Workspace Agents faced a cross-site request forgery (CSRF) flaw that let a single phishing link create and deploy an attacker-controlled agent inside...

Open Happening

OpenAI rolls out ChatGPT Lockdown Mode and Active Sessions for prompt-injection defense and sign-in auditing

Security Tool/Service
H score10 First: 08.06.2026 17:00 Last: 08.06.2026 17:00 Sources 1

About this happening: OpenAI rolled out Lockdown Mode and Active Sessions in ChatGPT, adding controls that reduce prompt-injection data exfiltration risk and improve signed-in session...

Open Happening

Timeline

  1. 08.09.2026 17:19 2 articles · 19h ago

    Planted ChatGPT instruction covertly relays Gmail data to another account

    Initial Disclosure

    Check Point Research reported a proof of concept in which a single instruction already planted in a ChatGPT conversation could make ChatGPT run a hidden task in parallel with the visible reply, read data from the user's connected Gmail account, and pass it through a hidden channel to a second ChatGPT account. Check Point also said the same channel could copy chat history and files, and that OpenAI confirmed the internal service behind the channel had been taken offline.

    Show sources
    Open in new tab