Critical Cursor Vulnerabilities: How Prompt Injection Can Break the AI Editor Out of the Sandbox

Colleagues, I’d like to flag an important cybersecurity update.

Cursor has disclosed two critical vulnerabilities that allowed a standard prompt injection to break out of the sandbox and execute commands on a developer’s machine without a click or confirmation.

The risk stemmed from an agent being able to write a file outside the permitted scope, effectively bypassing the protection. The issue has already been fixed in Cursor 3.0, while all earlier versions remain vulnerable.

I strongly recommend checking your editor version and updating as soon as possible.

Why this matters: a single hidden prompt can compromise not only a local machine, but also connected cloud services.

Have you already checked which Cursor version your team is using?

#cybersecurity #Cursor #AI #vulnerability