I have no idea how this is supposed to work, though. How can they guarantee that the agent does not see the password and username in a browser they fully control?!

There needs to be an OpenBao like implementation. I really like OpenBao but wish it was a little simplier to update policies.

Give an AI agent access to my 1Password vault? Sure, why not, what could possibly go wrong?