Oh, wow. I hadn't considered this angle: "There's also the risk that conversations between an attorney and client could become fair game in legal proceedings; a New York federal judge in February ordered a criminal defendant to provide prosecutors with documents he created for his lawyers because it already had been shared with a third party, which was Anthropic's Claude." https://www.fastcompany.com/91571498/ai-notetaker-work-meetings-privacy-data
@buherator @lrhodes that makes perfect sense and is completely insane at the same time 🤯
@buherator @lrhodes now I'm wondering what it means for AI integretion in knowledge based system (obsidian, confluence, notion)
@buherator @lrhodes yeah of course, but until these kind of AI agent tools, it always felt like they where sharing space for you that they couldn't access, same as if you rent a safe in a bank.
@buherator @lrhodes I mean: if you have a space on obsidian, confluence or notion, it feels like you're renting space, but they are not supposed to have access to the content. With AI added to it, it flips the equation. Not using AI agents within these tools so I never thought about it before.
@buherator Not unless you're using Gemini to summarize. AFAICT, the ruling applies specifically to prompts tendered by the defendent and output produced by the LLM. If Gmail or Gdocs are only conveying messages between the client and their lawyer, that's not covered by this ruling.
Basically, the premise is that the prompt is addressed to the LLM provider, not the lawyer, while the output is the work of a third-party, even if it summarizes priviliged material.