Why you shouldn’t use AI browsers like Atlas or Comet with logins to email, SharePoint or any other online service right now—even though that’s a major part of what makes an AI browser interesting. (Simon Willison’s Weblog)
Summary
- Prompt injection: A sneaky security risk where malicious instructions hide in websites, emails to trick AI agents.
- OpenAI's CISO Dane Stuckey addresses this head-on, calling it an "unsolved security problem" that adversaries will keep probing.
- To protect users, OpenAI is rolling out "logged-out mode" and "watch mode" to limit agent access, but admit the challenge remains.
