Choosing an AI Tool: A Due-Diligence Checklist for Firms

Most AI tools look alike in a demo. The differences that matter to a law or accounting firm sit in the data terms, the admin controls and the contract. This checklist gives you the questions to ask and what a clear answer looks like.

Why the questions matter

Both professions expect you to vet the vendors who handle client information.

For lawyers, ABA Formal Opinion 512 applies earlier guidance on outsourcing and cloud computing to AI tools. It suggests checking that the tool is configured to protect confidentiality, that the vendor’s obligations are enforceable, that you will be told about a breach or a demand for client data, and what the vendor retains during and after the relationship. It also says lawyers should read the tool’s terms of use and privacy policy, or rely on someone who has.

For accountants, the AICPA Code says that before disclosing confidential client information to a third-party service provider, a member should either have a contract requiring the provider to keep it confidential, with reasonable assurance about the provider’s safeguards, or get the client’s specific consent. A separate interpretation asks members to make sure a provider has the right qualifications and resources before using it.

Send the same questions to every vendor in writing. Keep the answers. They become part of your record that you took reasonable steps.

1. Training on your data

• Does the vendor, or any AI model provider it uses, train or improve models on your prompts, files or outputs?
• Is that the default, or does it depend on a setting someone could change?
• Does feedback, such as a thumbs-up button, change the answer?

A clear answer names every model provider involved and says what each one’s terms allow. “We take privacy seriously” is not an answer.

2. Retention and deletion

• How long are prompts, files and outputs kept?
• Can your firm set a shorter period?
• When you delete something, how long until it is gone from backups?
• Do any logs, such as abuse-monitoring logs at the model provider, keep content after you delete it?

Retention should match your firm’s records policy. Shorter is not always better, since you may need records for a matter. What matters is that you choose.

3. Where data is processed

• In which countries is data stored and processed?
• Does that differ by AI model or feature?
• Can you limit the firm to models processed in a specific country?

Some clients, contracts and engagements restrict where data may go. Know the answer before you accept work with those terms.

4. Subprocessors

• Which outside companies handle your data? Think hosting, model providers, email, support tools and analytics.
• Is the list published?
• Will you be told before a new subprocessor is added?

An AI tool is often a layer on top of other providers. Your data’s protection is only as clear as the weakest link you know about.

5. Access controls

• Can you control who gets an account, or can anyone sign up with a firm email address?
• Is two-factor authentication available? Can you require it?
• Does the tool support single sign-on with your identity provider?
• Can you set roles, so only some people can change settings or see reports?
• When someone leaves, how quickly can you remove access, and what happens to their data?

6. Audit logs and reporting

• Can an owner or admin see who used the tool and when?
• Is there a record of settings changes?
• Can you export logs if you need to investigate an incident or answer a client?

7. Isolation from other customers

• Is your firm’s data kept separate from other customers’ data? How?
• Can anyone outside your firm see your content through sharing features?
• Can your staff share content outside the firm, and can you turn that off?

8. Vendor access

• Which vendor employees or contractors can see your content?
• For what reasons? Support, abuse review, legal demands?
• Will the vendor tell you about a subpoena or other demand for your data, where the law allows?

9. Export and exit

• Can you export your data in a usable format?
• If you cancel, how long until the vendor deletes it, and will it confirm in writing?
• Does the vendor claim any rights in your inputs or outputs?

10. Certifications: what they do and do not mean

Certifications and audit reports are useful evidence. They are not a guarantee, and none of them makes your firm compliant with its professional rules.

SOC 2. A SOC 2 report is an independent CPA firm’s examination of a vendor’s controls, measured against the AICPA’s Trust Services Criteria for security, availability, processing integrity, confidentiality and privacy. It is a report, not a certificate. A Type 2 report tests whether controls worked over a period of time, which tells you more than a report on design alone. Ask for the report itself, check which product and which criteria it covers, check the date, and read any exceptions the auditor found.

ISO/IEC 27001. This standard sets requirements for an information security management system. Certification means an auditor found the vendor’s system for managing security risks conforms to the standard. Check the scope statement. A certificate may cover one office or one product, not the service you are buying.

HIPAA. HIPAA is a federal law, not a certification. The Department of Health and Human Services says it does not endorse or recognize private “certifications” of Security Rule compliance. If you handle health information, the question is whether the vendor will sign a Business Associate Agreement and what it covers.

No certification. A vendor without these reports is not automatically unsafe, and one with them is not automatically right for you. Without them, rely more heavily on the written answers above and on the contract.

11. Contract terms and data processing addendum

• Does the contract say the vendor will keep your data confidential and use it only to provide the service?
• Is there a data processing addendum, or equivalent terms, that you can sign?
• Does it require breach notice, and how fast?
• What are the limits on the vendor’s liability?
• Can the vendor change its terms without notice?

Read the documents that will govern the relationship, not the marketing page. If the privacy page and the contract disagree, the contract usually controls.

12. Cost model

• Is pricing per seat, per use, or both? Is there a seat minimum?
• What happens when you exceed usage limits?
• Is there a free trial, and does it come with the same data terms as the paid plan?
• Which models and features are included at each price?

For law firms, pricing also affects billing. Opinion 512 says a general-purpose tool that works like office equipment is normally overhead, and should not be charged to clients without advance disclosure. A per-use charge for one client’s matter can ordinarily be billed at actual cost.

Putting it together

Score each vendor against the twelve areas. You do not need perfect answers everywhere. You need answers you understand, that fit your clients’ expectations, and that you can explain if asked. Then record the decision in your AI policy and set a date to check again, because vendor terms change.

How Zrico fits

Each firm gets its own private workspace and databases. Claude is the default and is processed only in the US. GPT and Gemini are off unless the firm owner turns them on, and their processing is not limited to the US. No AI provider trains on prompts, files or answers. Sign-in is invite-only with two-factor (required for admins). Owners and admins get a usage and audit report and retention settings, and backups are encrypted nightly. Zrico has no SOC 2, ISO or HIPAA certifications. It costs $39 per seat per month (5-seat minimum, 30-day free trial). It does not make a firm compliant, and output must be reviewed by a professional.

Sources

• ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
• Model Rule 5.3, Responsibilities Regarding Nonlawyer Assistance: https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_5_3_responsibilities_regarding_nonlawyer_assistant/
• AICPA Code of Professional Conduct (see 1.300.040, Use of a Third-Party Service Provider, and 1.700.040, Disclosing Information to a Third-Party Service Provider): https://pub.aicpa.org/codeofconduct/ethicsresources/et-cod.pdf
• AICPA, SOC 2 guide (Reporting on an Examination of Controls at a Service Organization Relevant to Security, Availability, Processing Integrity, Confidentiality, or Privacy): https://www.aicpa-cima.com/cpe-learning/publication/soc-2-reporting-on-an-examination-of-controls-at-a-service-organization-relevant-to-security-availability-processing-integrity-confidentiality-or-privacy
• ISO/IEC 27001:2022, Information security management systems: https://www.iso.org/standard/27001
• HHS, Are we required to “certify” our organization’s compliance with the standards of the Security Rule?: https://www.hhs.gov/hipaa/for-professionals/faq/are-we-required-to-certify-our-organizations-compliance-with-the-standards/index.html
• OpenAI Platform documentation, Data controls in the OpenAI platform (example of model-provider abuse-monitoring retention): https://developers.openai.com/api/docs/guides/your-data

General information, not legal advice. Check the current rules in your jurisdiction.