Key takeaways
- ABA Formal Opinion 512 expects lawyers to read and understand an AI tool's terms of use and privacy policy, or consult someone who has[1].
- Ask the same questions of every vendor: retention, training on inputs, access, subprocessors, security, breach notice and exit.
- Voice tools raise biometric issues; Illinois BIPA damages run to $1,000 per negligent and $5,000 per intentional or reckless violation[2].
- Score answers in writing and put the important ones into the contract.
What due diligence means for an AI tool
Vetting an AI vendor is a professional duty as well as good business practice. ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure of, or access to, client information[3]. ABA Formal Opinion 512 adds that, as a baseline, lawyers should read and understand the terms of use, privacy policy and related contractual terms of any generative AI tool they use, or consult a colleague or external expert who has, and may need IT or cybersecurity help to understand them[1].
The opinion also points to earlier guidance on outsourcing that applies to AI providers: reference checks and vendor credentials, the vendor's security policies and protocols, its hiring practices, confidentiality agreements, its conflicts check system where relevant, and the availability of a legal forum if the vendor breaches the agreement[1]. That list is a sound outline for your questionnaire.
The core question set
| Area | Question | A good answer looks like |
|---|---|---|
| Retention | How long do you keep our inputs, outputs, recordings and transcripts? | A stated period you can configure, with deletion on request and at contract end |
| Training use | Are our inputs or outputs used to train or improve models, yours or anyone else's? | No, stated in the contract, not just a settings page |
| Access | Which of your staff can see our data, and when? | Named roles, logged access, only for support with our permission |
| Subprocessors | Which third parties process our data, and where? | A published list with notice before changes |
| Security | What independent security assessments do you have? | A recent third-party report you will share under NDA |
| Breach notice | How fast will you tell us about a security incident? | A fixed time in the contract and a named contact |
| Isolation | Can one customer's data appear in another customer's output? | A clear explanation of how data is separated |
| Exit | What do we get back if we leave, and when is everything deleted? | Full export in a usable format and written confirmation of deletion |
Retention and training use deserve the most attention
The two questions that most affect confidentiality are how long the vendor keeps your data and whether it uses your data to train models. A tool that learns from inputs can, in principle, surface information from one user's matter in another's output, which is why Opinion 512 requires informed consent before client information goes into such self-learning tools[1]. California's State Bar guidance likewise says lawyers must not input confidential client information into a tool that may present material risks to confidentiality or security without informed client consent[4].
- Get the answer in the contract or data processing terms, not only in a help article that can change.
- Check whether the answer differs between consumer, business and enterprise plans.
- Ask whether human reviewers at the vendor ever read inputs for quality or safety, and under what conditions.
- Ask what happens to data in backups after deletion, and how long backups are kept.
Security questions
- Is data encrypted in transit and at rest, and who controls the keys?
- Do you support single sign-on and enforce multi-factor sign-in for our users?
- Can we set roles so staff see only what their job requires?
- Do you keep an audit log of user and administrator actions that we can export?
- Where are your data centers, and can we choose to keep data in the US?
- When did you last have an independent penetration test, and what did you fix?
Voice tools and biometrics
AI phone answering, call transcription and voice analysis tools raise a separate issue: biometric data. Illinois' Biometric Information Privacy Act defines a voiceprint as a biometric identifier and bars private entities from collecting one without first informing the person in writing and obtaining a written release[5]. A prevailing party may recover liquidated damages of $1,000 per negligent violation or $5,000 per intentional or reckless violation, or actual damages if greater[2]. Ask every voice vendor directly whether its system creates voiceprints or other biometric identifiers, and get the answer in writing. Recording a call is not the same as creating a voiceprint, but you need the vendor to say which it does.
Source: Illinois General Assembly, 2025[2]
Show the data
| Item | Value |
|---|---|
| Negligent violation | $1,000 |
| Intentional or reckless violation | $5,000 |
State privacy law exposure
Consumer privacy laws can also apply to the firm itself. From January 1, 2025, a for-profit business falls under the CCPA if, among other tests, its annual gross revenue exceeds $26,625,000[6], and statutory damages for a data breach rose in 2025 to between $107 and $799 per consumer per incident, or actual damages if greater[7]. Even if your firm is below the thresholds, your vendor contract should match the obligations you would face if it were not, because client expectations follow the law's direction.
Red flags
- The vendor will not say whether inputs are used for training, or the answer differs by plan and you cannot tell which plan you are on.
- No list of subprocessors, or a refusal to notify you before changes.
- Security claims with no independent report available.
- A contract that lets the vendor change data terms by updating a web page.
- Marketing that claims the tool gives legal advice or performs like a lawyer. The FTC has acted against exactly that claim[8].
- No clear answer on deletion at the end of the contract.
Score and decide
Send the questionnaire
Use the same written questions for every vendor and give a deadline.
Score each answer
Use a simple scale: meets the requirement, partly meets it, does not meet it or no answer.
Set deal-breakers
Decide in advance which answers disqualify a vendor, such as training on client inputs or no breach notice commitment.
Check references
Speak to at least one law firm of similar size using the product, and ask about support and incidents.
Contract the answers
Move the key answers into the contract or data processing terms before signing.
Contract terms to ask for
- No use of firm data to train or improve models, for any plan or feature you use.
- Defined retention periods and deletion on request and at termination, with written confirmation.
- Breach notification within a fixed number of hours or days, with a named contact.
- Notice before adding or changing subprocessors, with a right to object or exit.
- Your right to export all data, recordings and transcripts in a usable format.
- A legal forum and governing law you can realistically use.
Extra questions for AI intake and phone tools
- How does the system disclose to callers that they are speaking with AI, and can we set the wording?
- Can we approve the exact intake questions and the wording used for matters we do not handle?
- What does the system say if a caller asks for legal advice, and can it ever give it?
- How are urgent calls escalated, and what happens if nobody answers the escalation?
- Are calls recorded, how is consent handled, and can we change the recording disclosure by state?
- Who at the vendor can listen to recordings or read transcripts, and why?
Who should run the review
In a small firm, the review usually falls to the managing lawyer, with help from the firm's IT provider for the security questions. If nobody in the firm can assess a security report, pay for a short review by someone who can, rather than relying on the vendor's summary. Record who reviewed what and when. That record is part of showing the reasonable efforts the confidentiality rule expects.
Keep the review proportionate. A drafting tool used only for marketing copy needs a short check; a tool that records every client call or holds every matter file needs the full questionnaire, a security report and contract changes before anyone signs.
Example: vetting an AI phone answering vendor
Example (fictional): a five-lawyer firm in Illinois evaluates two AI phone answering services. Both record calls and produce transcripts. The firm sends its questionnaire, and one vendor confirms in writing that it does not create voiceprints, does not train on customer data and deletes recordings after a retention period the firm sets. The other says data handling is "covered in our privacy policy", which allows training for product improvement. The firm treats the second answer as a deal-breaker, signs with the first, and adds the no-voiceprint and no-training answers to the contract.
Re-vet on a schedule
Vendors change terms, ownership and subprocessors. Re-check each AI vendor at least yearly, and immediately after an acquisition, a major product change or a security incident. Keep the completed questionnaire and the date of each review with your approved-tools register so the next person can see what was checked and when.
Frequently asked questions
What should a law firm ask an AI vendor first?
Whether your inputs are retained and for how long, and whether they are used to train models. Those two answers decide most confidentiality questions.
Do lawyers have to read an AI tool's terms of use?
ABA Formal Opinion 512 says lawyers should read and understand the terms of use and privacy policy, or consult someone who has[1].
Do AI phone tools create biometric data?
Some voice tools may create voiceprints, which Illinois BIPA treats as biometric identifiers requiring written notice and release[5]. Ask the vendor in writing.
Is a vendor's privacy policy enough?
Not on its own. Put the key commitments, such as no training on your data and deletion at exit, into the contract.
References
American Bar Association, Standing Committee on Ethics and Professional Responsibility (2024). Formal Opinion 512: Generative Artificial Intelligence Tools (archived copy) (opens in a new tab) โ
Illinois General Assembly (2025). Biometric Information Privacy Act (740 ILCS 14) (opens in a new tab) โ
American Bar Association (2025). Rule 1.6: Confidentiality of Information, ABA Model Rules of Professional Conduct (archived copy) (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
Illinois General Assembly (2025). Biometric Information Privacy Act (740 ILCS 14) (opens in a new tab) โ
California Privacy Protection Agency (2025). Updated Monetary Thresholds in CCPA (opens in a new tab) โ
California Privacy Protection Agency (2025). Updated Monetary Thresholds in CCPA (opens in a new tab) โ
Federal Trade Commission (2025). FTC Finalizes Order with DoNotPay That Prohibits Deceptive 'AI Lawyer' Claims (opens in a new tab) โ