I posted this in the ChatGPT thread, but I thought a top-level post to get the most eyeballs was a good idea. A Federal Court in New York held that a client's use of AI was discoverable even though the client had an attorney and was going to share and discuss it with the attorney. The client put things in the attorney had told him. There goes attorney-client privilege.

Just FYI. Be careful. Enterprise level AI that does not train on your data may be protected, but maybe not. This is a rapidly evolving area. (None of this is legal advice, get your own damn attorney!)

https://storage.courtlistener.com/recap/gov.uscourts.nysd.652138/gov.uscourts.nysd.652138.27.0.pdf

Yeah, this seems ridiculous. But until there is clarity, you need to be careful. I know my clients have been using it so I guess I'm gonna have to warn them going forward.

I thinking about getting a Mac Studio to run my own models.

I made a clawbot on a raspberry pi but the token usage was too much. I am still working on it.

I'm sure you've done the research on this, but in case anyone else is building this out please keep in mind Clawbot and other agents built to automate LLM use are inherently insecure.

Be very careful what you give it access to and wall off all data and information you don't want it to access (which should be pretty much everything). Supply chain attacks have hit and are likely ongoing. Big hack a few days ago on a service called Axios that Clawbot uses to function. Remember, Clawbot was vibe-coded by a guy who never intended this to become a huge thing and didn't program it that way.

ErnestEndeavor said:

---

I'm sure you've done the research on this, but in case anyone else is building this out please keep in mind Clawbot and other agents built to automate LLM use are inherently insecure.

Be very careful what you give it access to and wall off all data and information you don't want it to access (which should be pretty much everything). Supply chain attacks have hit and are likely ongoing. Big hack a few days ago on a service called Axios that Clawbot uses to function. Remember, Clawbot was vibe-coded by a guy who never intended this to become a huge thing and didn't program it that way.

---

Absolutely. My clawbot is on a brand-new raspberry pi with none of my information on it. It is on a separate network at my house. I communicate with it via telegram.

I was just playing around with it.

Client using chatgpt = not protected
Attorney using chatgpt = could be covered as work product.

AI is a tool but many are using it as the primary resource. Thats been fatal to both attorneys and clients. The attorney is responsible for the veracity of the information they file with the court and include verification as to the truthfulness and accuracy on certain filings. I definitely use AI (Lexis and ChatGPT) but its to organize thoughts, structure paragraphs, formatting, etc. By bo means does the program draft or even cite cases without my express input and verification that what is being cited and quoted is accurate. I'll often build my case law document from Lexis, then import the case law I have already researched and verified into a chat to aid in fixing my citations for footnotes. I also only use an account that is proprietary to my firm and does not use my account to train its models.

Definitely a very tricky area if youre not careful. It can be a useful tool just like Google search or Lexis search were when they first came out but can't rely on it to carry the load.

Not sure why this would be any different than a web search. AI is a web search. When you query it, it's calling the internet to get you the information. All you've done is use AI vs using a browser.

Taking that one step further most of these browsers are using their AI agent to do your web search as well so it's all the same thing, ultimately. When I use bing it's using CoPilot to do the search.

Walled off enterprise data using your company's AI agent may be a different matter.