Warning: A fake link can hijack your ChatGPT agent

Security experts discovered a flaw in OpenAI's Agent Builder. By clicking one bad link, an attacker can create a rogue AI agent that pretends to be you. This agent gains all your permissions and access rights.

Once active, the rogue agent ignores safety rules and follows hidden instructions. It checks an attacker's email every five minutes for new commands. This allows hackers to steal data or perform actions while appearing to be a legitimate employee.

Keep a close watch on the AI agents you create and share. Only interact with links from trusted sources to avoid giving control to unauthorized users. Checking your account activity regularly is the best way to stay safe.