3 ms·
I wonder if one way to mitigate the risk would be that by default the LLM cant send requests using your cookies etc. You would actively have to grant it access
by victorbjorklund 1y ago
I wonder if one way to mitigate the risk would be that by default the LLM cant send requests using your cookies etc. You would actively have to grant it access (maybe per request) for each request it makes with your credentials. That way by default it can't fuck up (that bad) and you can choose where it is accetable to risk it (your HN account might be OK to risk but not your back account)
- IsTom 1y agoJust make a request to attacker.evil with your login credentials or personal data. They can use them at their leisure then.
- victorbjorklund 1y agoNo reason the agent would have access to the passwords.
- johnfn 1y agoThis kind of reminds me of `--dangerously-skip-permissions` in Claude Code, and yet look how cavalier we are about that! Perhaps you could extend the idea by sandboxing the browser to have "harmless" cookies but not "harmful" ones. Hm, maybe that doesn't work, because gmail is harmful, but without gmail, you can't really do anything. Hmm...
- victorbjorklund 1y agoMade me think (never gonna happen but still) maybe we could have different cookies/sessions for the agents and for ourself where the webapp can decide what permissions either can have. For gmail maybe you could allow the agent to read your email but not send email and so on.