Consider an ordinary working situation. A paralegal at a small law firm is trying to turn a rough set of notes into something a client can read before a meeting. She opens an AI writing tool she already uses for personal projects and pastes in the content, which includes a client’s name, some financial figures, and a short account of a dispute that hasn’t been resolved. The firm has never told her whether that kind of information is allowed in that tool, so she makes the call on her own and gets back to her afternoon.
AI tools are fast and familiar, already open in a browser tab on a work computer, so the point at which a clear limit should apply is easy to miss. The firm never sat down and decided what can and cannot go into them, so the decision falls to whoever is working on the file that afternoon.
The employee is left to decide what the firm never did
The paralegal in this example has no policy to break. She’s doing what any employee does when a task needs finishing and a tool is already open on the screen, and professional services firms encourage this kind of initiative every day while saying little or nothing about where AI fits into it.
Knowingly breaking a known rule is a bad choice, and the person who broke it has to own it. Working inside a firm that never set a rule at all puts the ownership somewhere else. It sits with firm leadership, since the staff member filling the space the firm left open never had a real alternative to choose from. Many leadership teams only discover this once something has already gone wrong.
Submitted information can travel further than staff expect
Once information is typed into an AI tool, what happens to it depends on the product and the account behind it. OpenAI states plainly in its enterprise privacy commitments that it does not train its models on submitted data by default across its ChatGPT Business and Enterprise plans. That protection is tied to business accounts. A personal or free account used for the same work can sit under different default settings.
No single platform is the focus here. A firm using AI without oversight typically doesn’t know which tools its people are using or what they are entering into them. It often doesn’t know what type of account sits behind the work either, and that detail determines how the information gets handled downstream. Telling staff to use good judgment doesn’t close any of that, because good judgment depends on information the firm hasn’t provided.
That uncertainty compounds in professional services. The information typed into these tools is often tied to an active client matter, carrying details a client assumes are staying inside the firm. The AI tool has no way of recognizing any of that unless someone decided, in advance, that it needed protection.
Client trust depends on where information ends up
Law firms and accounting firms hold information clients expect to stay inside the relationship. A client file or a set of financial records is the kind of material people share because they trust the firm to keep it contained. Professional obligations back that trust with real consequences.
The American Bar Association addressed this directly for lawyers in its 2024 opinion on generative AI tools. The opinion ties AI use to the existing duty of confidentiality under Model Rule 1.6 and recommends reviewing a tool’s privacy terms before any client information goes into it. The responsibility for that review sits with the firm and the individual lawyer.
Accountants carry a comparable duty. The AICPA’s Confidential Client Information Rule, set out in its Code of Professional Conduct, requires specific client consent before confidential information is disclosed to a third party, and entering client data into an AI tool can count as exactly that kind of disclosure. Neither obligation was written with AI specifically in mind. Both apply regardless.
Financial services firms sit under a similar expectation. FINRA reminded member firms in 2024 that its existing supervision and communication rules apply to generative AI use exactly as they apply to any other technology and named the protection of customer information as a central risk area for firms to manage. That obligation doesn’t disappear because the information passed through a new kind of tool first.
A single mistake doesn’t have to end a client relationship. The deeper risk is a firm losing track of where its information goes and losing the ability to answer a client clearly if they ever ask where theirs went. In a market built on referral and reputation, a client who senses their information was handled carelessly often says nothing formal at all. They stop referring new work instead, and that erosion is far harder to trace back to its source than a formal complaint would be.
Governance needs to catch up with how AI is already being used
AI spreads through a firm’s daily work before anyone sits down to decide how it should be governed, and that sequence is where this risk begins.
The NIST AI Risk Management Framework places governance first, calling on organizations to assign clear ownership and accountability for AI decisions before anything else gets built on top of them. That is what the previous article in this series covered, including choosing approved tools and setting clear rules for what staff can put into them.
A short, clear standard does most of the work here. Naming which tools are approved and setting limits on what can be typed into them covers the core of it, and reinforcing that standard regularly keeps staff from drifting back to personal habits out of convenience. A clear AI policy for employees removes that uncertainty by setting the rule in advance, and it turns AI compliance from a reactive scramble into a standing part of how the firm already operates.
Banning AI is not the useful response here. Staff have already found real value in these tools, and a well-run firm should be able to capture that value without leaving confidential information to whoever happens to be at the keyboard that day. Governance exists to make that possible. It is how a firm decides in advance what its people should never be left to work out alone.
