Resource
What law firms need to know about AI tools and client confidentiality.
A federal court has already ruled that using a consumer AI tool can destroy attorney-client privilege. Not because of a data breach, but because of what the tool's terms of service allowed. The risk isn't hypothetical, and it doesn't require anything to go wrong for it to be real.
The case that changes how this conversation should go.
In United States v. Heppner, a federal court ruled that an attorney's use of a consumer AI tool destroyed privilege over the communication. There was no breach. Nothing was hacked. The privilege was lost because the AI tool's terms of service permitted data retention, model training, and third-party disclosure. And that possibility alone was enough.
This is the detail most firms haven't absorbed yet. The conversation about AI and confidentiality usually centers on what happens if something goes wrong. Heppner shows the exposure exists the moment the tool is used, regardless of outcome. The question isn't whether the AI tool will be breached. It's whether the tool's own terms already created the exposure before anyone typed a word.
What Rule 1.6 actually requires.
Confidentiality under Rule 1.6 requires reasonable efforts to prevent unauthorized disclosure of client information. The ABA's guidance on AI makes clear this isn't a new rule invented for AI. It's the same obligation applied to a new category of tool, and the "reasonable efforts" standard has to account for what a given AI tool's terms of service actually say about retention, training, and third-party access.
This creates a real obligation before a tool is ever used, not after. A lawyer who hasn't reviewed an AI vendor's data handling terms hasn't met the threshold for reasonable efforts, even if nothing ever goes wrong with that specific use.
There's a second, less obvious obligation running alongside confidentiality: competence. The ABA has been explicit that lawyers must understand both the capabilities and the limitations of the AI tools they use, and that this is an ongoing obligation as the tools themselves keep changing. Avoiding AI entirely isn't a safe default either. At least one body of guidance suggests that refusing to use available tools that would benefit a client's matter could itself raise a competence question. The standard isn't "don't use AI." It's "understand what you're using well enough to use it responsibly."
The adoption gap that already exists at your firm.
Recent industry data shows the large majority of legal professionals are already using AI tools in their work, while a substantial share of firms have no formal governance policy in place. That gap is almost certainly already true inside your firm, whether or not anyone has looked.
This isn't a hypothetical risk that might emerge someday. It's a description of where most firms already are right now: associates and paralegals using AI tools for research, summarization, and drafting, often on personal accounts or free consumer versions, with no review of what those tools' terms of service actually permit. The firm's policy, if one exists, usually hasn't caught up to what's actually happening on individual devices.
The risk specific to firms with multiple practice areas.
This is the part of the AI confidentiality conversation that doesn't get said often enough, and it matters more for law firms than almost any other kind of business.
Traditional ethical walls and conflict screens were built around a world where information stayed compartmentalized by matter, by attorney, by practice group. AI systems weren't designed with those boundaries in mind. Even a firm that builds or licenses its own AI tool for internal use can create a problem if one attorney inputs client-confidential information and the system surfaces related information to another attorney working on a different, potentially adverse matter. The conflict screen that exists on paper doesn't automatically exist inside the AI system unless someone has specifically built it that way.
For a firm representing multiple clients in the same industry, or handling matters where conflicts of interest are a live concern, this is a governance question that has to be answered before an AI tool is adopted firm-wide. Not discovered after the fact.
Supervision doesn't stop with the attorney who typed the prompt.
Managing partners and supervisory attorneys have an affirmative obligation to establish clear policies on permissible AI use and to make reasonable efforts to ensure that both lawyers and non-lawyer staff comply with them. This extends the confidentiality obligation beyond the partners who are thinking about it to paralegals, legal assistants, and administrative staff who may be using AI tools without any awareness that ethical obligations apply to them too.
And the obligation doesn't get delegated away. An associate directed by a partner to use a particular AI tool remains personally responsible for complying with the rules of professional conduct, regardless of who told them to do it. "I was told to use it" is not a defense if the tool itself created the exposure.
One more obligation worth naming: billing. If AI use produces a meaningful time savings on a matter, billing a client the full hourly rate for time that wasn't actually spent raises a separate ethical issue around reasonableness of fees. This is a smaller piece of the overall AI governance conversation, but it's a concrete obligation that's easy to overlook when the focus is entirely on confidentiality and privilege.
Jurisdiction matters more than it used to.
More than thirty states have issued AI-specific guidance for attorneys, and the specifics vary meaningfully. Some states require explicit disclosure of AI use in court submissions. Some require continuing legal education credits specifically in AI competency. Some impose stricter rules around cross-jurisdictional data handling when an AI vendor processes information outside the state where the matter is being handled.
A firm practicing across state lines doesn't get to pick the easiest standard and apply it everywhere. The governance policy has to account for the jurisdictions the firm actually practices in, not just the one where the firm is headquartered.
What governance actually looks like for a law firm.
The same three-part approach that applies to AI governance generally applies here, with the legal-specific obligations layered on top.
Inventory what's actually running. Before any policy gets written, find out what AI tools are currently in use across the firm. Including tools individual attorneys and staff have adopted on their own, browser extensions, and AI features embedded in software the firm already uses. Most firms are surprised by what this turns up.
Evaluate each tool against the confidentiality standard, not just convenience. For every tool in use or under consideration, the question is specific: what do the terms of service say about data retention, model training, and third-party access. A tool with a zero-data-retention agreement is a meaningfully different proposition than a free consumer tool that trains on user inputs by default. And the difference is the kind of thing Heppner turned on.
Build a policy that accounts for ethical walls, not just data handling. For firms with multiple practice areas or potentially adverse client relationships, the governance process has to specifically address whether an AI tool can surface information across matters that should remain separated.
Extend it to everyone, not just the attorneys who asked for it. Paralegals, legal assistants, and administrative staff need to understand the policy applies to them, and the firm needs a way to verify that it's actually being followed, not just distributed.
Most firms don't know what's already running.
Before any of this can be addressed, a firm needs an honest answer to a basic question: what AI tools are actually in use right now, across every device and account, by everyone at the firm. Most firms have never had that answer with any confidence.
What changes when this is handled correctly.
The firm knows what's running before a court ruling or a bar complaint forces the question. Tools in use have been evaluated against the specific terms that matter for privilege, not adopted because they seemed convenient. The policy accounts for ethical walls between matters, not just general data handling. Staff at every level understand the policy applies to them. And if a client, a court, or a bar inquiry asks what the firm's AI governance looks like, there's a documented answer instead of an assumption.
If you're not confident in what's actually running across your firm or whether your current approach holds up against Rule 1.6 and the competence standard, that's exactly what an IT Environment Review is for.
Guides for law firms
Take something with you.
A first-steps Copilot guide for legal teams, plus an incident response plan template you can fill in for your firm.
AI visibility
Want to see what AI tools are actually running in your firm?
Most firms find more tools in use than they expected, including tools embedded in software they already pay for. We can show you exactly what's running before you build a policy around assumptions.
Common questions
Questions leadership usually asks first.
Continue reading
Related work and reading.
IT for Law Firms
How we manage identity, DMS, backup, and AI tool governance for firms held to Rule 1.6 and client scrutiny.
Read more: IT for Law FirmsAI Readiness & Governance
Inventory the tools already in use, evaluate them against confidentiality obligations, and get an approval process in place.
Read more: AI Readiness & GovernanceGovernance, Risk & Compliance
Document the controls behind Rule 1.6, competence, and supervisory obligations instead of assuming they exist.
Read more: Governance, Risk & ComplianceNext step
Schedule an IT Environment Review.
Get a clear picture of what AI tools are actually running in your firm and whether your governance approach holds up against Rule 1.6 and the competence standard.