Executive Summary
An AI incident feed lists a zero-click flaw in ChatGPT and Codex integrations that enabled GitHub OAuth token theft through indirect prompt injection, and EchoLeak, a zero-click Copilot flaw that exfiltrated enterprise documents.
What Happened
This news feed aggregates recent AI security incidents, including a zero-click vulnerability in ChatGPT and Codex integrations that enabled GitHub OAuth token exfiltration via indirect prompt injection. It also references EchoLeak, a zero-click Copilot flaw that allowed automated exfiltration of sensitive enterprise documents.
Affected Products
- OpenAI ChatGPT
- OpenAI Codex
- Microsoft Copilot
Vulnerability / Attack Technique
Zero-click indirect prompt injection leading to token and document theft
Technical Mechanism
Untrusted content processed by the assistant carried instructions that caused it to send OAuth tokens or documents to attackers without user interaction.
AI Security Category
LLM01:2025 Prompt Injection
Part of the Prompt Injection topic.
Relevant Frameworks
RealGround Analysis
Zero-click means the user never has a chance to notice. RealGround analysis: assistants with connected accounts need output and egress controls, not just input filtering.
Recommended Controls
- Scope OAuth tokens given to AI integrations
- Restrict assistant egress to approved destinations
- Apply vendor fixes for affected integrations
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
PrivacyScrubber: https://privacyscrubber.com/news/
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.