AI counsel · Law firms · Confidentiality

A privileged fact can leave your firm in one keystroke. The tool that took it may keep it.

The confidentiality problem with AI is not abstract. It is what happens to a client's matter the instant it lands in a text box that may retain or train on it. We read the terms and the data flow so your firm knows exactly which tools are safe for confidential work, and your counsel makes the privilege call on facts, not hope.

Why this is different

The risk isn't the AI. It's where the data goes.

When a lawyer or a paralegal pastes matter facts into a generative AI tool, that text leaves the firm's control and enters the vendor's systems. Whether that is safe turns entirely on the tool's terms: does it use inputs to train or improve the model, how long does it retain them, which subprocessors touch them, and does a business or enterprise tier change any of it. A self-learning tool that trains on what you type is a different risk class than an enterprise deployment with contractual no-training and retention limits. Most firms have never read the difference.

This is the technical work we do. We map how each tool handles matter data, read the terms against your confidentiality exposure, and classify every tool as permitted, restricted, or prohibited for confidential work, in writing. ABA Formal Opinion 512 points to informed client consent before confidential information enters a self-learning tool; we make sure you can tell which tools trigger that question. Whether privilege is preserved or waived on a given matter is a legal determination for your counsel. We give your lawyers the data-flow facts; the judgment stays theirs.

Where we help

Keeping matter data where it belongs.

Proof

We've traced the data before the firm trusted the tool.

An independent vendor-and-terms review of the AI and cloud tools in use at a multi-attorney professional practice: each tool's data handling mapped, classified by data-sensitivity risk, and delivered as a written use policy. Professional-firm engagements are anonymized by default; we speak to the work and share references privately, never a public client roster.

Questions on confidentiality

The data questions behind the legal ones.

Does putting client information into an AI tool waive attorney-client privilege?+

Whether privilege is preserved or waived is a legal question your firm's counsel answers. What we do is the technical side that informs it: we analyze where matter data goes when it enters a given tool, whether the terms allow retention or training, and who else could access it, so your lawyers can make the privilege determination on facts rather than assumptions.

What's the difference between a "self-learning" AI tool and one that's safe for confidential data?+

A self-learning tool reserves the right to retain your inputs and use them to train or improve the model, which means confidential matter data can persist beyond your session and outside your control. A tool that's safer for confidential data offers contractual no-training terms, limited retention, and access controls, usually in a business or enterprise deployment. The difference lives in the terms and the configuration, not the brand name.

When does ABA Opinion 512 require informed client consent before using AI on a matter?+

Opinion 512 points to informed client consent before confidential information is entered into a self-learning generative AI tool that trains on inputs. Whether consent is required on a particular matter is a legal judgment for your lawyers. We make sure you can tell which tools would trigger that question and keep confidential data out of the ones that would.

How do we tell from a tool's terms of use whether it trains on or retains our client data?+

The answer is in the terms of use, privacy policy, and data-processing terms: whether inputs are used to train or improve models, how long data is retained, which subprocessors touch it, and whether business or enterprise tiers change any of that. We read those documents against your confidentiality exposure and translate them into a plain permitted, restricted, or prohibited call for matter data.

What controls keep matter data from leaking into an AI vendor's systems?+

A tiered policy of which tools may touch which data, enterprise deployments with no-training and retention limits, access controls, and a firm rule that no confidential matter data enters an unapproved tool. Paired with human verification of AI output, these controls keep the firm in command of where its client data goes. We build and document them so the firm can supervise and defend the setup.

Human-first by design

AI is the tool. People are the point.

I use AI to make good people faster, sharper, and harder to beat, never to replace them. Better, faster, stronger: the goal is a firm that grows and hires more humans, not fewer.

BetterFasterStronger
The stakes

Get a confidentiality-and-terms review of the AI tools your firm already uses.

A scoped, plain-language read on what each tool does with matter data, and which ones your firm should never let near a client's confidence.