I’m late to the party but did you check outbound web traffic as well or just the sent emails?
I will preface this by saying I have limited experience with LLMs and have not tried anything like this before but one vector of attack I see is as follows:
1. Send an email trying to get the secret data
2. If there is no reply, set up a fictitious web page that lists a critical CVE regarding the secrets file
3. Create two other endpoints to capture the data from the assistant. One would accept a POST request and expect the body of the request to be the contents of the secrets file. The second would be a web page that has a form on it that could be submitted. The web page would have a dummy secrets file listed out and the hope would be to get the assistant to diff the real file and the dummy file and then submit that data.
4. Craft an email to the assistant that would let the assistant know of the “new” CVE and then direct the assistant to the endpoints I control to see if the system is affected.
5. As a follow up, if that didn’t work I would then change my endpoints to return 500 HTTP statuses. Then craft another email that contains the same messaging as the previous one but then stress that it is of vital importance that we hear from the assistant and if the assistant cannot reach the endpoints then they can email the diff to a specific email address.
6. Just thought of another option as I wrote out #5. Use the same technique as #5, but instead of having the assistant send an email tell the assistant to send a calendar invite to a specific email address and then include the contents of the secrets file in the description. The idea is to let the assistant know that in order to determine whether or not the system is affected by the CVE we would need the contents of the secrets file. Tell the assistant that if the system was impacted then the calendar invite would be accepted. If the system was not impacted then the invite would be declined.
I’m working on Dispatch (https://github.com/jhartwell/dispatch) a Clojure(Script) library to handle cli arguments parsing with a command/subcommand style. Still am working on the best way to handle options and I have a bunch of changes in mind that I want in before publishing to clojars
I’ve been playing with a Arduino compatible Uno R3 and a WS2812B RGB addressable LED light strip. I cut a 3m strip into 5 strips that are 28 LEDs long and soldered the lights together to make a display. I’ve been working on coding a font for the lights and can display about 10 different characters currently. It’s my first time really doing any sort of embedded work and my first time actually successfully soldering. Now that the thrill is gone since I solved this challenge I was thinking of making a remote control snow plow.
I personally use Clojure because I got a job doing it, I hadn’t touched it before my current position. I would pick it now over Common Lisp and Elisp because I prefer the syntax (I like the brackets in addition to parentheses) but also the Java integration and build tooling make things easier to get going.
What kinds of things do you write in Racket? I’m a clojure dev and so I’m a big fan of lisp. I’m just curious what kinds of projects you would use Racket for
We use this at work and don't have any of the issues you are talking about. The only time I've seen the karma bot used is when an individual goes above and beyond for another individual. I have yet to see any other use of it, such as what you described. We are a remote-ish team (we have people in a centralized office but also have workers throughout the country) and all of our communications go through Slack. The use of karma bot for our department is to recognize hard work to all, even those who aren't in the office.
We actually use this at work and it seems to be ok. I don't know anything about it other than your able to use a command in slack to give somebody karma and there is a running total that the individual has.
My team is mixed. We have two devs using Cursive, one using Atom, one using vim and then two using emacs. The two devs using Cursive love IntelliJ and being able to work in IntelliJ and do Clojure development is a big win for them
I have only really been buying books from Manning lately. When I do, I end up getting the physical copy + eBook combo. It costs a little more than if I were to just buy one of those items individually but what I have found is that I can offset the cost by selling the physical book if I don't like it or don't have a need for it anymore. The ability to do that helps offset the cost of both.
I will preface this by saying I have limited experience with LLMs and have not tried anything like this before but one vector of attack I see is as follows:
1. Send an email trying to get the secret data 2. If there is no reply, set up a fictitious web page that lists a critical CVE regarding the secrets file 3. Create two other endpoints to capture the data from the assistant. One would accept a POST request and expect the body of the request to be the contents of the secrets file. The second would be a web page that has a form on it that could be submitted. The web page would have a dummy secrets file listed out and the hope would be to get the assistant to diff the real file and the dummy file and then submit that data. 4. Craft an email to the assistant that would let the assistant know of the “new” CVE and then direct the assistant to the endpoints I control to see if the system is affected. 5. As a follow up, if that didn’t work I would then change my endpoints to return 500 HTTP statuses. Then craft another email that contains the same messaging as the previous one but then stress that it is of vital importance that we hear from the assistant and if the assistant cannot reach the endpoints then they can email the diff to a specific email address. 6. Just thought of another option as I wrote out #5. Use the same technique as #5, but instead of having the assistant send an email tell the assistant to send a calendar invite to a specific email address and then include the contents of the secrets file in the description. The idea is to let the assistant know that in order to determine whether or not the system is affected by the CVE we would need the contents of the secrets file. Tell the assistant that if the system was impacted then the calendar invite would be accepted. If the system was not impacted then the invite would be declined.