Executive Summary
Researchers disclosed a critical flaw in ChatGPT Workspace Agents that could let a phishing link create, authorize and deploy an autonomous agent inside an organization, plus data-exfiltration risks involving conversations and uploaded files.
What Happened
The Hacker News reports that researchers disclosed a critical vulnerability in OpenAI ChatGPT Workspace Agents that could allow a phishing link to create, authorize, and deploy an autonomous agent inside an organization. The same source also reports research describing ChatGPT data-exfiltration risks involving sensitive conversation content and uploaded files.
Affected Products
- OpenAI ChatGPT Workspace Agents
Vulnerability / Attack Technique
Phishing link that deploys an authorized agent
Technical Mechanism
A crafted link could create and authorize an agent in the victim's workspace, giving the attacker an agent with organizational access.
AI Security Category
LLM06:2025 Excessive Agency
Part of the AI Agent Excessive Agency topic.
Relevant Frameworks
RealGround Analysis
Agent creation is now a privileged action an attacker can target. RealGround analysis: workspaces should restrict who can create and authorize agents.
Recommended Controls
- Restrict agent creation and authorization to admins
- Review existing workspace agents for unknown ones
- Train users on agent-authorization phishing
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
The Hacker News: https://thehackernews.com/search/label/OpenAI
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.