Your associates already have an AI policy. You just didn't write it.

Ask a managing partner whether the firm uses AI and the answer is almost always the same. Not really, we haven’t rolled anything out. Ask a few more questions and it turns out an associate is using ChatGPT to draft a demand letter, a paralegal is summarizing deposition notes with an AI note taker, and someone in billing found a tool that cleans up invoice language. Nobody at the firm approved any of this. It just happened.

New research backs up what that conversation always reveals. PagerDuty’s survey, published in June 2026, found that 66% of office workers have used an AI tool for work while believing their employer didn’t allow it. At organizations with 1,500 or more employees, that climbs to 72%. At smaller organizations, it’s still 60%. There’s no reason to think a law firm is the exception.

The confidentiality problem hiding in that number

For most businesses, this is a data governance issue. For a firm, it’s a professional responsibility issue, and the two look very different.

Forty three percent of workers admitted entering emails or other work content into a public AI tool. More than a third put customer information in. Read that as client information for a firm handling anything from a demand letter to a settlement summary. Thirty one percent uploaded financial data or confidential documents, which for a firm can mean matter files, trust account records, or a client’s business strategy.

That’s the duty of confidentiality, and it doesn’t pause for AI. ABA Model Rule 1.6(c) requires a lawyer to make reasonable efforts to prevent unauthorized disclosure of information relating to a client’s representation, whether the disclosure is a lost laptop or a paste into a chatbot. ABA Formal Opinion 512, issued in July 2024, applies this directly to generative AI: confidentiality, competence, and supervision obligations all travel with the tool, and the lawyer stays responsible for the output regardless of what wrote it.

For California firms, the same duty runs through a different set of rules. Business and Professions Code section 6068(e) sets the underlying confidentiality obligation, and California Rule 1.1 on competence, backed by Formal Opinion 2010-179, extends that duty to how the firm uses technology. The State Bar’s own guidance on generative AI, issued in November 2023, points the same direction: confidentiality doesn’t pause because the tool is convenient.

Why associates aren’t asking permission first

A third of AI users said they’d deliberately avoid telling a supervisor they used it. About 30% said the policy felt too restrictive or they worried how coworkers would react. Another 29% said they weren’t sure what the rules were.

That’s not defiance. That’s confusion wearing a guilty face, and at a firm it’s sitting on top of a client’s privileged information.

Here’s the part that should bother a managing partner more: 86% of workers say their employer already has an AI policy, and 81% believe leadership operates under a different set of rules than everyone else. If associates think the policy bends for partners, they’ll treat it as optional for themselves. Some go further and switch to a personal device specifically so nobody at the firm can see what they’re doing with a client’s file.

A ban won’t hold, so build something that will

Most workers, 72%, believe they understand AI better than whoever’s managing it where they work. At billion-dollar companies, that figure is 80%. Associates who are confident they’re right don’t stop using a tool because a memo told them to. They just stop mentioning it.

Sending a client’s facts into a public AI tool isn’t that different from emailing the same file to a personal account to finish over the weekend. It feels like a convenience. It’s a disclosure the rules don’t treat as harmless, and it’s one your firm can’t see happening.

The fix is a sanctioned option, not a memo banning everything. Microsoft 365 Copilot, kept inside the firm’s own tenant, gives associates a fast drafting tool without sending client data to a public model. Pair it with two paragraphs of clear rules, not twelve pages, naming what never goes into any AI tool regardless of which one: client identifying facts, matter files, anything under a protective order or NDA. Enforce it the same way for partners as for first year associates. The 81% who assume leadership gets a pass are watching for exactly that inconsistency.

Back to the question this started with

Could you say today, with confidence, that no client information has gone into a tool the firm never approved? If the honest answer feels closer to “probably not” than “yes,” that’s the gap worth closing before a regulator, an insurer, or opposing counsel is the one asking. We help firms figure out what’s actually being used, put a workable policy in place, and pick tools that keep client data inside the firm’s control. It’s usually less work than partners expect, and considerably less than cleaning up after an inadvertent disclosure.