A law firm associate pastes a client’s confidential memo into a general-purpose chatbot to get a quick summary. The summary is good. The problem is that the memo now lives somewhere nobody at the firm can control, and the client’s secrets may have just walked out the door. That scenario is no longer hypothetical, and it is pushing bar associations, legal ethicists, and court systems to figure out what attorney-client privilege even means when the intermediary is a large language model.
The Maryland Daily Record recently turned a spotlight on this exact question, and it is worth paying attention beyond the legal profession. Privilege is one of the oldest protections in the legal system, shielding candid communication between lawyers and clients so that advice can be given honestly. It was built for a world of paper files, locked cabinets, and phone calls. AI tools that ingest, summarize, draft, and search documents complicate every one of those assumptions.
The core tension is simple to state and hard to resolve. When a lawyer uses a third-party AI service, does that count as sharing confidential information with an outsider? Courts have long held that disclosing privileged material to a third party can waive the protection, with narrow exceptions for agents working under the lawyer’s direction. Whether an AI vendor qualifies as such an agent depends on contracts, data handling, and how much control the firm actually retains. Those details are often buried in terms of service that nobody reads closely.
There is a second layer that gets less attention. Many AI tools improve by learning from the data users feed them. If a firm’s inputs get absorbed into a model’s training pipeline, the confidentiality problem compounds, because the information may resurface in outputs to entirely unrelated users. Enterprise agreements increasingly promise not to train on customer data, but the legal world has learned the hard way that vendor promises need verification, not faith.
Law firms are responding in uneven ways. Some have banned public chatbots outright. Others have built private instances of models behind their own firewalls. A growing number have adopted formal AI use policies that spell out which tools are approved, what data can go in, and who reviews outputs before anything reaches a client. Legal ethics opinions in several jurisdictions have started to sketch the boundaries, generally agreeing that competence now includes understanding the risks of the technology you choose to use.
Clients have a stake here too, and many do not realize it. If you hire a lawyer who runs your case files through an AI system that mishandles them, the privilege protecting your communications could be compromised, and you may never be told. Some corporate legal departments have started asking outside counsel pointed questions about their AI practices during vendor reviews. That pressure from the client side may end up doing more to standardize behavior than any ethics rule.
The courts are watching as well. Judges have begun issuing standing orders requiring parties to disclose whether AI was used in preparing filings, and a few have imposed sanctions when lawyers submitted AI-generated material containing fabricated citations. Those cases are about accuracy rather than privilege, but they signal a broader judicial skepticism toward lawyers who treat AI as a black box.
None of this means AI has no place in legal work. Document review, contract analysis, and research are genuinely well suited to these tools, and firms that use them carefully can serve clients faster and cheaper. The point is that convenience is not the same as safety, and the profession’s traditional guardrails were not designed with probabilistic software in mind.
The practical takeaway for anyone outside the legal field is that this is a preview of a problem heading for every industry that handles sensitive information. Health care, finance, journalism, and human resources all depend on confidentiality in some form, and all of them are watching employees paste private material into chatbots right now. The legal profession is simply the first to confront the question with a centuries-old doctrine on the line. How it answers will shape the playbook everyone else follows.