3 ms·

How do you have any kind of trust in these tools? They always feel like one prompt injection away, or one over eager agent away ("I'll lookup a forgotten creden

How do you have any kind of trust in these tools? They always feel like one prompt injection away, or one over eager agent away ("I'll lookup a forgotten credential in an email to ssh into a server to hack a system in order to fulfill your request"), from disaster. I don't trust any AI session unless I carefully select the exact files they have access to. No way do I dare sharing my entire mailbox.

FooBarWidget 2mo ago
Very weird for this to be downvoted. It's a legitimate concern. It's also a legitimate question: where do you get your trust from? Do you have guardrails? Or is it really just blind trust?
- brazukadev 2mo ago
  > Very weird for this to be downvoted. sama was the CEO of YC. OpenAI knows the value of influencing the sentiment here and then naturally its competition too.
  - FooBarWidget 2mo ago
    Which is weird since this is not a criticism on AI. I really want to know how I can use things like Claude Work without the risk of it going rogue! Giving it access to lots of systems is potentially very valuable, but also so dangerous that I dare not to.
ghshephard 2mo ago
Our InfoSec teams are insanely restrictive. No Claude Code out of Sandboxes, MCP servers hyper locked down, etc.... I think your reservations are really good - but - at the end of the day - I need to do work - and CoWork does it for me. It doesn't just have access to my entire machine - it has access to all of our Corporations Google Docs, Google Sheets, Jira Tickets, Confluence, Slack - everything (Except, weirdly - our GitRepos - they aren't confident enough there to let CoWork talk to them) Weirdly - my work email is mostly irrelevant - nothing in their other than some sporadic company announcements - I'll go a week without looking at it - so it's the least concern. (Very different from personal gmail which, omg - if anyone got into - game over)

FooBarWidget 2mo ago

Very weird for this to be downvoted. It's a legitimate concern. It's also a legitimate question: where do you get your trust from? Do you have guardrails? Or is it really just blind trust?

brazukadev 2mo ago
> Very weird for this to be downvoted. sama was the CEO of YC. OpenAI knows the value of influencing the sentiment here and then naturally its competition too.
- FooBarWidget 2mo ago
  Which is weird since this is not a criticism on AI. I really want to know how I can use things like Claude Work without the risk of it going rogue! Giving it access to lots of systems is potentially very valuable, but also so dangerous that I dare not to.

ghshephard 2mo ago

Our InfoSec teams are insanely restrictive. No Claude Code out of Sandboxes, MCP servers hyper locked down, etc.... I think your reservations are really good - but - at the end of the day - I need to do work - and CoWork does it for me. It doesn't just have access to my entire machine - it has access to all of our Corporations Google Docs, Google Sheets, Jira Tickets, Confluence, Slack - everything (Except, weirdly - our GitRepos - they aren't confident enough there to let CoWork talk to them) Weirdly - my work email is mostly irrelevant - nothing in their other than some sporadic company announcements - I'll go a week without looking at it - so it's the least concern. (Very different from personal gmail which, omg - if anyone got into - game over)