AI is genuinely useful on a case file: timelines, contradictions, cross prep. It's also a confidentiality landmine if you feed client material into the wrong tool. The line between the two is knowable.
Your duty of confidentiality applies to AI uploads exactly as it applies to any other disclosure. The question to ask of any tool: does the provider train on what I upload, and do they retain it? Free consumer chatbots often answer yes to both. Paid business tiers usually answer no in writing.
Generally defensible with client material: Claude's paid and Team or Enterprise plans, ChatGPT Team and Enterprise, Microsoft 365 Copilot commercial, and Google Workspace's Gemini and NotebookLM. Confirm the terms of YOUR tier; commitments differ by plan, and a protective order can restrict AI use regardless.
Even with a safe tool, discovery arrives as scanned PDFs, video, and audio that a chatbot can't digest. Asking good questions is the easy part. The slow part is converting forty files into text the AI can read, then knowing what fits in one conversation.
That conversion is our AI reference folder add-on: your whole case, discovery and transcripts, as clean text files with an index, token counts, a one-file version for a single chat, and a page of starter prompts written for criminal defense work. It ships with a one-page advisory version of this article, so the safe-tools question is answered on paper, in your file.
Drop the folder into the AI your firm already pays for and start asking.
AI reference folder, $50 with any order