AI, Confidentiality, and Client Data: What an M&A Attorney Should Require From Any Platform

Why the duty of confidentiality applies to the platform

When an attorney uploads a client's financial statements into a platform, the information may be disclosed to or accessed by a third party. That is not new in principle. Firms have used document management systems, data room providers, and cloud email for years. What is different with language-model tools is that the platform may read documents to generate teasers, CIMs, and buyer matches. Reading is the feature. The attorney therefore needs to know what the platform does with what it reads. The ABA's guidance treats the risk as dependent on the client, matter, task, and tool, and calls for reasonable efforts against unauthorized disclosure or access. [1]

The practical questions are where the data is stored, who at the vendor can access it, whether the vendor's models are trained on it, whether outputs for one customer can draw on another customer's inputs, and what happens when the engagement ends. The attorney should get answers in writing, in the vendor's terms of service or a separate data agreement, before the firm's first client is onboarded. A verbal assurance from a sales representative is not an answer. The firm should also consider whether its engagement letter should disclose that it uses a software platform and explain what that platform does with the client's data. If informed consent is required for the tool and use at issue, a generic boilerplate clause may not be enough. [1]

Isolation: one client's data should never touch another's

The first requirement is isolation. The firm will run multiple clients on the same platform, and the platform may serve other firms. A client's data room, questionnaire answers, valuation inputs, and buyer outreach record should be inaccessible to any other client, any other firm, and any process that generates outputs for someone else.

Isolation matters most in buyer matching. A platform that builds a prospect list by reading the client's data room and website is using the client's information for that client's process. A platform that also draws on what it has learned from other sellers' data rooms creates a different confidentiality risk. The attorney should ask the vendor directly: when the platform generates a prospect list for my client, what inputs does it use, and are any derived from other customers' confidential data?

Isolation also matters for the data room itself. Buyers who sign NDAs receive staged access to specific documents. The platform should enforce that staging and log every access. The attorney should be able to see who opened what and when, because that log can be important evidence if a confidentiality dispute arises later.

No training and traceability

The second requirement is that the client's data is not used to train models that serve other customers. A platform that improves its CIM generation by training on every CIM it has produced could carry forward the substance of one client's business description into the model that writes the next client's. Even if no sentence is copied, the attorney needs to know what use is permitted and how information may surface. The vendor's answer should be unambiguous and contractual: client data is not used to train models for other customers. If a vendor cannot say that, the attorney should treat the use as unresolved rather than rely on a sales explanation.

The third requirement is traceability. A valuation range that the attorney will discuss with the client, and that may be referenced in an IOI comparison or LOI negotiation, must be explainable. The attorney should be able to point to the inputs, adjustments, and method, and should get the same output from the same inputs every time. A valuation engine that produces different numbers on different days from the same financial statements is not a tool the attorney can rely on without further review.

The same standard applies to the CIM. Every factual statement in the book should trace to a questionnaire answer, a data room document, or the client's approval. The attorney edits the CIM and the client approves it. A platform that generates plausible text without showing its sources shifts the burden of verification entirely onto the attorney. Deterministic output and source tracing make a generated document reviewable rather than merely readable. The attorney should test both before trusting either. NIST recommends contracts and vendor assessments that address data privacy, security, content provenance, and records of changes made by third parties. [2]

Export and deletion at the end of the engagement

The fourth requirement is that the firm can export the complete record at any time and can require deletion when the engagement ends. The record includes the data room, questionnaire, generated documents and revisions, buyer list, outreach log, NDA record, IOIs and LOIs as received, and client approvals. This is the firm's file, and the firm's retention obligations apply to it. A platform that cannot produce it in a usable form, or that retains it after the firm has asked for deletion, is a platform the firm should not use for client work. The contract should state what is retained, for how long, and how deletion is verified. [1]

Export also protects the firm against vendor failure. If the platform ceases to exist mid-engagement, the firm must be able to continue the process with what it has. The attorney should test an export before relying on the platform for a live matter, including whether revisions, permissions, timestamps, and access records remain usable.

Security questions and a test for this week

Alongside the AI-specific questions, conventional security questions still apply. Where is data stored, and in what jurisdiction? Is it encrypted in transit and at rest? Who at the vendor can access it, under what controls? What is the vendor's incident notification commitment? Does the vendor undergo independent security assessment, and will it share the results? Can the firm enforce multi-factor authentication for its own users and for the client's portal access? These questions are not unique to AI platforms, but they should be asked of this vendor before this client's data is uploaded. NIST identifies third-party assessment, approved provider lists, incident contingencies, and fallbacks as useful controls for third-party generative AI systems. [2]

Before the next client is onboarded, the responsible partner should write the four requirements above as questions, send them to the vendor, and file the written answers in the firm's vendor management record. If the firm has already onboarded clients, the same exercise should be done now. An attorney who can produce the vendor's written data commitments when a client asks what happens to financial statements has an answer. An attorney who cannot has a problem. The firm should confirm its own position with its own counsel on how applicable professional rules and other legal requirements apply to the activities it takes on and the fees it charges.

How Acquivo Pro fits

Acquivo Pro is designed around the requirements described here. Client data is isolated and not used to train models for other customers. The valuation engine is deterministic: the same inputs always produce the same output, and figures trace to their source. The platform reads the data room, company website, and questionnaire to build a prospect list tailored to that client, and generates teasers and CIMs from the client questionnaire and data room. The client sees a branded document portal under the firm's name where disclosures are approved before anything goes out. Software prepares, organizes, matches, and tracks; the attorney and the client decide and act. To walk through a process on a client you have in mind, request a demo through the Contact page.