00:00:00–00:00:05
Your AI can read that page.
Should it obey it?

00:00:05–00:00:10
Data can contain instructions.
Treat the page as untrusted content.

00:00:10–00:00:15
Limit which tools it can use.
Limit what they can do.

00:00:15–00:00:20
Require approval before
consequential actions execute.

00:00:20–00:00:25
Layers reduce risk.
They do not guarantee safety.