Key takeaways
- ABA Formal Opinion 512 says lawyers should understand a generative AI tool's terms of use and privacy policy, or consult someone who has.
- Opinion 512 says boilerplate engagement letter language is not sufficient informed consent for self-learning AI tools.
- In United States v. Heppner (S.D.N.Y. 2026), a public AI platform's privacy policy helped defeat privilege and work product claims.
- California's State Bar guidance says not to input confidential client information into AI tools lacking adequate confidentiality and security protections.
- An AI vendor contract is usually several documents, and consumer and enterprise plans can carry different training and retention terms.
- The clauses to check first are training, retention, human review, legal process, subprocessors, security, breach notice and exit.
Before an attorney puts client information into an AI tool, ABA Formal Opinion 512 expects someone at the firm to understand the tool's terms. Those terms arrive as a stack: a services agreement, a data processing addendum, a privacy policy, product-specific terms and a list of subprocessors, and together they decide who can see client information after it leaves the firm. This checklist walks through how a law firm can review an AI vendor contract for privilege and confidentiality risk before the tool touches a matter.
Can an AI vendor's terms affect attorney-client privilege?
Yes. A court may read the provider's terms to decide whether anyone could reasonably expect the communication to stay confidential, and terms that allow retention, training and disclosure to third parties work against that expectation.
In United States v. Heppner (S.D.N.Y. Feb. 17, 2026), Judge Jed Rakoff held that documents a criminal defendant created with a public AI platform, on his own and without his lawyers' direction, were protected by neither attorney-client privilege nor the work product doctrine. The court relied on the platform's privacy policy, which said user inputs and outputs could be retained, used to train models and disclosed to third parties, including government authorities. The case involved a consumer account used by a client acting alone. It did not decide how firm-managed tools under negotiated confidentiality terms should be treated, which is exactly why the contract deserves a close read.
What does ABA Formal Opinion 512 require before you use an AI tool?
ABA Formal Opinion 512 says lawyers should read and understand an AI tool's terms of use, privacy policy and related contractual terms, or consult someone qualified who has, before client information goes in. California's State Bar guidance gives similar direction for California firms.
Opinion 512 (July 29, 2024) also says a client's informed consent is required before information relating to the representation goes into a self-learning tool, and that general, boiler-plate engagement letter language is not sufficient. If no one at the firm can evaluate the terms, the opinion points lawyers to a colleague or outside expert who can.
California firms also answer to state law. Business and Professions Code section 6068(e)(1) requires an attorney "to maintain inviolate the confidence, and at every peril to himself or herself to preserve the secrets, of his or her client." The State Bar's practical guidance on generative AI says a lawyer "must not input any confidential information of the client into any generative AI solution that lacks adequate confidentiality and security protections," should review the terms of use to determine how the product uses inputs, and should confirm the provider does not use inputs to train or improve its product.
Which documents make up an AI vendor contract?
An AI vendor contract is usually five or six documents, and the one your attorney clicked through may not govern. Collect all of them before you start.
- The master or enterprise services agreement, including its order of precedence clause.
- The data processing addendum or data protection terms.
- Product-specific or AI-specific service terms.
- The privacy policy and any acceptable use policy.
- The subprocessor list, including any third-party model provider.
- Security documentation, such as a SOC 2 Type II report.
Check which plan the terms apply to. Consumer, team and enterprise plans from the same provider can carry very different commitments on training and retention.
What clauses should a law firm check in an AI vendor contract?
Check the twelve clauses that decide who can see, keep or disclose client information, and record the exact language and section number for each.
- Training: a written commitment that inputs, outputs and uploaded files are not used to train or improve models, including through feedback or rating features.
- Retention and deletion: a defined retention period, deletion at the firm's request and at termination, and how long content is kept for abuse monitoring or in backups.
- Human review: when, if ever, vendor staff or contractors can view firm content, and under what controls.
- Legal process: a commitment to notify the firm before responding to a subpoena or government request where the law allows, and to limit disclosure to what is legally required.
- Subprocessors: who processes the data, including the underlying model provider, with notice of changes and a right to object.
- Confidentiality scope: a definition of confidential information that clearly covers prompts, outputs and uploaded documents.
- Security: encryption, sign-in through the firm's identity provider with multifactor authentication, exportable audit logs and data location.
- Breach notice: a specific notification deadline and the information the vendor will provide.
- Ownership: the firm keeps ownership of inputs and outputs, with no license to the vendor beyond providing the service.
- Matter permissions: the tool honors document management permissions and ethical walls and does not search beyond them.
- Unilateral changes: whether the vendor can change terms by posting a new version, and how much notice the firm gets.
- Exit: export of firm data and certified deletion when the contract ends, timed so the firm can still meet its own client file retention duties.
How should a firm document the review?
Keep a short review file for each AI tool that shows what the firm checked, who checked it and when. Opinion 512 and the State Bar's guidance both expect managerial and supervisory lawyers to set clear policies on permitted uses of generative AI and to train lawyers and staff.
- The version and date of every document reviewed.
- Who reviewed it, including any ethics counsel or IT adviser consulted.
- The settings configured to match the terms, such as retention, training opt-out and sign-in.
- The firm's decision on client consent and any matter-specific limits.
- A date to re-review at renewal or when the vendor changes its terms.
What should a law firm do this quarter?
Start with the AI tools attorneys already use, then make vendor review a standing step in software intake.
- List every AI product attorneys and staff use, including features inside existing software.
- Collect the current terms for each and run them through the checklist above.
- Move client work out of any consumer accounts whose terms permit training or broad disclosure.
- Ask ethics counsel to review your consent approach under Opinion 512 and California authority.
- Add AI vendor review to your intake process for new software.
How s90 helps
s90 handles the technical side of AI vendor review for law firms. We read the security documentation, confirm how retention, training and logging actually behave, connect sign-in to your identity provider and set data loss prevention rules that keep privileged material out of unapproved tools. Your ethics counsel makes the legal calls. If your attorneys are asking for a new AI tool, we are glad to help you review it before it reaches a matter.
