Can a prosecutor’s office use AI with case files?

By Verillian. Published . Updated .

Yes, within the confidentiality and supervision duties ABA Formal Opinion 512 describes, and CJIS rules where the file holds criminal justice information.

Two attorneys review documents together at a desk.

Yes, a prosecutor’s office can use AI with case files, and two sets of rules attach. The rules of professional conduct bind every prosecutor as a lawyer, and ABA Formal Opinion 512 applies the ABA Model Rules to generative AI.

The CJIS Security Policy binds the office as an entity whenever the file includes criminal justice information. Meeting one does not meet the other.

What does ABA Formal Opinion 512 say about case information going into a GAI tool?

ABA Formal Opinion 512, issued July 29, 2024 by the Standing Committee on Ethics and Professional Responsibility and based on the ABA Model Rules of Professional Conduct, says a lawyer using a generative AI tool must be mindful of the Model Rule 1.6 duty to keep confidential all information relating to the representation. Our post on small firms covers what it asks of a private practice.

It singles out self-learning tools, which it describes as learning from themselves as they cull more data, for a specific risk. Information entered for one matter can surface later in response to other users’ prompts.

For such a tool, the opinion says the client’s informed consent is required before inputting information relating to the representation, and a boilerplate engagement-letter clause is not enough to obtain it.

Model Rule 1.1’s competence duty adds a second requirement. A lawyer needs a reasonable understanding of the capabilities and limitations of any tool they use. The opinion does not mention prosecutors or government lawyers, and it is the prosecutor’s own state rules of professional conduct that bind, read with that state’s own ethics opinions.

The opinion is written around a lawyer’s duties to a “client,” which is why a prosecutor-specific reading matters here.

Integrating AI: Guidance and Policies for Prosecutors, from the National Best Practices Committee convened by the Prosecutors’ Center for Excellence, applies the confidentiality duty directly to that setting. It reaches the facts, evidence, witness information, and other details of a criminal case.

Before using a GAI tool to draft a case memo, or to analyze phone and text message data obtained with a search warrant, a prosecutor has to determine whether inputting that material could reveal it to the tool’s developer or other users, or divulge data that is legally under the control of the issuing court.

Do CJIS’s security rules reach a prosecutor’s office the same way they reach police?

Yes, wherever the office has access to criminal justice information. The CJIS Security Policy’s scope reaches all entities with access to CJI or that operate systems used to process, store, or transmit it, and the PCE guide ties a prosecutor office’s CJIS obligations to keeping its access to NCIC and other CJIS databases.

The guide names the consequence directly. The FBI writes CJIS guidelines but does not accredit programs or services, so even in a seemingly secure product, introducing a new GAI tool requires reevaluating its CJIS compatibility.

A GAI component that downloads data into another server to process it, or a program that retains a file to train itself, could present a potential CJIS violation.

Our post on CJIS audit requirements for AI use covers the audit rules. The policy’s scope turns on access to CJI, not on whether the system is a legal-research or drafting tool rather than a records system.

What does confidentiality mean specifically for victim information?

It means preventing disclosure of a victim’s personal identifying information, such as an address, a date of birth, or a Social Security number. That is a separate question from general case confidentiality.

The PCE guide points to state victim-rights legislation that requires prosecutors to protect that information, and says prosecutors must ensure AI is not used in a manner that would allow its disclosure.

Those statutes vary by state, so check your own state’s victim-protection law directly rather than assume a Rule 1.6 analysis covers the same ground. The two obligations run in parallel, and a tool that clears one has not automatically cleared the other.

Who is responsible for how a vendor’s AI tool handles case data?

The office’s managerial and supervisory lawyers, under the duties Model Rules 5.1 and 5.3 describe. Formal Opinion 512 says managerial lawyers must establish clear policies on permissible GAI use, and supervisory lawyers must make reasonable efforts to ensure compliance, including training for the lawyers and nonlawyers who use the tools.

For tools run by outside providers, it applies earlier ABA guidance on cloud computing and outsourcing.

That guidance suggests that lawyers ensure the tool is configured to preserve confidentiality and security, with an enforceable obligation and notice of any breach; investigate its reliability, security measures and policies; determine whether it retains submitted information or asserts proprietary rights to it; and understand that its servers can fail or be attacked.

The PCE guide applies the same duty to a prosecutor’s office. Vendors providing cloud storage or case-management software must be employing GAI in a manner that does not risk disclosure of confidential information, and contracts with AI vendors must conform with the attorneys’ professional obligations.

What should a prosecutor office’s AI policy require?

  • Run the Rule 1.6 risk analysis before case facts go into a tool, especially a self-learning or publicly available one. Where the analysis calls for informed consent, a boilerplate clause will not do, and who gives that consent for a prosecutor’s office is a question Opinion 512 does not address; it belongs with the office’s counsel and the state’s own ethics authorities.
  • Keep CJI out of public AI tools entirely, and reevaluate CJIS compatibility whenever a new GAI feature is introduced into a system that already touches CJI, rather than assuming compatibility carries over from the underlying software.
  • Separate victim protections from general confidentiality, and check the office’s own state statute, since the two duties do not automatically travel together.
  • Vet every vendor’s AI features under Rules 5.1 and 5.3, not only the vendor’s core product. Confirm retention, breach notice, and proprietary-rights terms in writing before case data reaches the tool.
  • Verify what a GAI tool produces before it is used or filed. The PCE guide tells prosecutors to independently verify any output. Formal Opinion 512 ties the amount of review to the tool and the task. A citation, a case summary or a draft memo is a starting point.
  • Do not expect AI to change who decides disclosure to the defense. Using AI on a case file does not change the rules that decide what goes to the defense, Brady and the jurisdiction’s own discovery rules; whether a given AI draft or output falls within them is a separate question that no published decision has settled as far as we can find. See our post on whether an AI-generated police report is discoverable for how that analysis runs on the report side of the same case.

Sources

  • American Bar Association, Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, Generative Artificial Intelligence Tools, July 29, 2024. Full opinion text, read November 3, 2026.
  • Prosecutors’ Center for Excellence, National Best Practices Committee, Integrating AI: Guidance and Policies for Prosecutors, January 2025. Full guide, read November 3, 2026.
  • CJIS Security Policy, version 6.1, June 25, 2026, FBI Criminal Justice Information Services Division. Full policy text, read November 3, 2026.

What record does Verillian keep?

Every AI request Verillian captures on an enrolled device is a signed record on your office’s own servers, each entry naming the person assigned to the enrolled device, the device and the time. It’s one input to the confidentiality risk analysis Formal Opinion 512 calls for.

What Verillian does for public safety agencies

See Verillian govern AI on a real device.