Post #4153081
2026-07-16 23:41 UTC
IMO this has problems but not for the reasons people are saying
https://1password.com/blog/1password-for-claude
If you use Claude a lot, you will eventually want to give it some credentials.
For 1Password, `op` (https://www.1password.dev/cli) already solved most of the problems. You can read secrets into the environment, or create sub-shells with the appropriate environment, all with opaque identifiers. It will invoke an auth dance / biometrics before proceeding
This pattern is similar, but... wronger
Replies (1)
-
@neilk@xoxo.zone 2026-07-16 23:45
Wronger because we shouldn't be giving an agent passwords when they have READ AND WRITE ACCESS TO THE WHOLE WEB. The attack surface is infinite and the LLM is unmonitored and is making its own decisions. It's been shown you can trick them just by sending extra instructions in *HTTP headers.* If you're really serious about automating web interactions with an LLM, you need to constrain exactly what it's allowed to visit (maybe on the network level) and give it a time-limited and scoped auth token