Caffeine AI, panopticon?

Developers
1

So I’ve been using Caffeine for a month now. It’s pretty impressive what it’s capable of but there’s one thing now happening that doesn’t make me feel very secure. I’m building a trading bot that according to the main Caffeine chat is okay to build, but every prompt I now make in that chat it replies with that it cannot process the prompt due to a security issue.

“:shield: Your prompt does not pass the security guard.”

The main chat stated: “No, your project is not being actively watched or monitored by admins. All projects on Caffeine are private to you unless you choose to share them or publish them. Admins do not review or oversee individual projects unless there is a specific support or security issue that requires attention. Your work and ideas remain confidential within the platform.”

So in essence it says it’s not being watched, unless… I cannot help but being reminded of a year or so back when users here on the forum were actively temporarily banned when there was a discussion about Dfinity leadership.

So, if my work remains confidential within the platform why is it suddenly after 130 versions into the project a security issue? Am I being monitored, and are they being monitored? Are projects being stolen before they can get launched?

I’d like to know more about this topic!

Cheers

3

First, am not Caffeine Support, just community trying to help.

That Guard is automatic and based on the prompt, not a thing of the project changing status.

As a proof, can you try some other prompt that is simple? Like, refactor my Spec or Frontend/Backend?

I hit that Guard already 2 times, and the Agent was right in both, it could get me into trouble.

Can you share what the prompt is exactly?

In my case, I was doing something with auth tokens, that clearly was a much better idea to just use Principal authentication. And another one was to do with changing Admin to be called something else, like Owner or something. Keeping consistency on Admin is a good security practice.

Looking forward to your prompt to support you better :+1:

4

In the meantime the issue seems to have resolved itself and Caffeine actually suggested to code the thing it initially rejected to do.

My prompts were rejected entirely, even if I would just say hello

1 Like