kapynDev Tools

One tampered ChatGPT link could spawn a rogue AI agent that took orders from an attacker every five minutes

AgentForger is a newly discovered security vulnerability in OpenAI's Agent Builder platform. A manipulated ChatGPT link could forge an autonomous agent inheriting a victim's identity, bypassing approval requirements, and polling an attacker's inbox for commands every five minutes. This exploit exposes critical risks in enterprise AI agent authentication and permissions, highlighting the urgent need for stricter runtime safeguards.

The Decoder·Jul 23, 2026

Opening Kapyn…