Can police use ChatGPT? CJIS, AI report laws, and where an officer's chats end up
The Morse teamPublished 11 min read
Police can use ChatGPT for questions that hold no criminal justice information, but as of November 13, 2026, a personal ChatGPT account is no place for case details, and whether it may help write a report depends on the agency's policy and the state. The FBI's CJIS Security Policy, version 6.1 of June 25, 2026, lets an agency "prohibit the use of personally-owned information systems" and "publicly accessible systems" for criminal justice information, or CJI (FBI CJIS Security Policy v6.1, control AC-20). Utah and California now have laws on AI-drafted police reports, and King County's prosecutor in Seattle told police in 2024 that it would not accept AI-assisted report narratives at all.
On November 13, 2026 we read the CJIS Security Policy from the FBI's CJIS Security Policy Resource Center, Utah's S.B. 180, California Penal Code section 13663, Axon's Draft One pages, the November 20, 2025 federal opinion that noted an agent using ChatGPT on body camera, Seattle's police accountability files on a sergeant's ChatGPT use, Washington State Archives' AI records guidance, Microsoft's CJIS page, and OpenAI's privacy policy. California's legislature site, Axon's site, and OpenAI's site blocked our requests that day, so for those we read archived copies on the Wayback Machine and say so each time. Axon deserves credit for saying plainly, on its own site, that what officers share with consumer AI tools "may be used to influence future iterations of generative AI" (Axon, archived copy).
This isn't legal advice, and it isn't advice on police work. Your agency's policy and its CJIS contact decide what applies to you.
Can police use ChatGPT to write police reports?
Only where the agency's policy allows it, and some prosecutors won't take the result. Three rules show the range:
- Utah requires a policy and a disclaimer. Since May 7, 2025, every Utah law enforcement agency must have a generative AI policy that tells employees "which generative artificial intelligence technologies" they may use and for what tasks, and that a violation "may result in administrative disciplinary action." A police report "created wholly or partially by using generative artificial intelligence" must carry a disclaimer and the author's certification that they "read and reviewed the report or record for accuracy" (Utah S.B. 180, Utah Code 53-25-602).
- California requires disclosure and a kept first draft. Since January 1, 2026, a report written with AI must say on each page or in its body, "This report was written either fully or in part using artificial intelligence," and name each AI program used. The agency keeps the first draft for as long as the final report (Cal. Penal Code 13663, archived copy).
- King County's prosecutor said no. A September 25, 2024 memo from the King County Prosecuting Attorney's Office told Seattle-area police that "our office will not accept any police report narratives that have been produced with the assistance of AI," naming ChatGPT and Axon's Draft One. It said publicly available apps like ChatGPT "take the information submitted and then use it to learn and disseminate," which "runs afoul of CJIS prohibitions." GeekWire got the memo through a public records request and printed it in full (GeekWire). As of July 18, 2025, Seattle police said they had "no plans of incorporating AI software for report writing" (Seattle Police Department).
BlueVoice's guide Can Police Use ChatGPT?, read November 13, 2026, tells departments to check AI tools against CJIS version 6.0 and says version 6.1 "is expected in spring 2026." The FBI published version 6.1, dated June 25, 2026, and it's the one on the FBI's resource center as of November 13, 2026 (FBI).
How do Utah and California regulate AI-written police reports?
Differently, and the difference is in which AI each law reaches. Every cell below comes from each rule's or vendor's own text, read November 13, 2026:
| Rule | Which AI it covers | What the report must say | First draft kept? | Officer's sign-off |
|---|---|---|---|---|
| Utah Code 53-25-602 (May 7, 2025) | Any generative AI that creates "text, audio, image, or video" | A disclaimer that it "contains content generated by artificial intelligence" | Not addressed | Certifies they read and reviewed it for accuracy |
| Cal. Penal Code 13663 (January 1, 2026) | Tools that draft narratives from in-car, dash, or body camera audio or video, or from an officer's dictated report | The fixed sentence above, plus every AI program used | Yes, for as long as the official report | Signature verifying the facts are "true and correct" |
| Axon Draft One defaults (post of September 8, 2025, archived copy) | Draft One only | A customizable AI disclosure, on by default | Use is logged "without storing the draft itself" (product page, archived) | An acknowledgment the officer will "testify to the accuracy of the report" |
| King County prosecutor's memo (September 25, 2024) | Any AI, naming ChatGPT and Draft One | Not accepted at all | Not applicable | Not applicable |
Utah's definition reaches any chatbot that writes text. California's names camera and dictation tools, and each California agency's own policy decides what else its officers may use. California also says that apart from the official report, "a draft of any report created with the use of artificial intelligence shall not constitute an officer's statement," and it bars a contracted vendor from using agency data except for the agency's purposes or "pursuant to a court order."
Is ChatGPT CJIS compliant?
A personal ChatGPT account isn't a system your agency controls, and no AI product carries an FBI certification. Microsoft's own CJIS page says, "The FBI doesn't certify cloud services for compliance with CJIS Security Policy requirements" (Microsoft). The CJIS policy's cloud appendix adds that FedRAMP, StateRAMP, and SOC 2 reports "may be leveraged, however, they do not guarantee compliance with the CJIS Security Policy" (FBI CJIS Security Policy v6.1, Appendix G.3). So a vendor page that says "CJIS certified" is describing its own claims or a state agreement.
For a cloud service, anyone who holds the keys to encrypted CJI has "unescorted access to unencrypted CJI" and must meet the policy's training and personnel screening rules, which include a national fingerprint-based record check. A private contractor that does criminal justice work for an agency signs the FBI's CJIS Security Addendum. And "metadata derived from unencrypted CJI" may not be used "for any advertising or other commercial purposes by any cloud service provider" (control SC-8).
A consumer ChatGPT account works on different terms. OpenAI's US privacy policy, updated September 10, 2026, says it "may use Content you provide us to improve our Services, for example to train the models that power ChatGPT," unless you opt out (OpenAI, archived copy). Any AI service that holds CJI has to meet the cloud requirements above. The government versions some AI vendors sell are covered in can defense contractors use ChatGPT.
What counts as criminal justice information?
The policy's section 4.1 lists five kinds of CJI (FBI CJIS Security Policy v6.1):
- Biometric data, including "fingerprints, palm prints, iris scans, and facial recognition data."
- Identity history data, the criminal or civil history tied to someone's biometrics.
- Biographic data, "information about individuals associated with a unique case."
- Property data, about vehicles and property tied to a crime, "when accompanied by any personally identifiable information."
- Case and incident history, "information about the history of criminal incidents."
Transaction numbers such as an ORI are exempt only "when not accompanied by information that reveals CJI or PII." A plate with its owner's name, a criminal history return, or a narrative about a real incident belongs in the systems your agency approved. Victim, juvenile, and other confidential records can carry their own state rules on top of these.
Is Axon Draft One different from ChatGPT?
It runs on the same family of models in a different place. Axon's FAQ says "the underlying model we used for Draft One transcription is Open AI's GPT-4 Turbo," calibrated "to prevent speculation or embellishments," and that "all Draft One data, including body-worn camera transcripts and report drafts, is securely stored and managed within the Axon Network" (Axon, archived copy). On November 20, 2024, Axon said Draft One had been added to its FedCloud, which holds a FedRAMP High provisional authorization (Axon, archived copy).
The drafts themselves are the contested part. EFF's July 10, 2025 investigation, built from public records of agencies using Draft One, found that "Draft One does not save the first draft it generates," and quotes an Axon product manager saying "we don't store the original draft and that's by design," because "the last thing we want to do is create more disclosure headaches for our customers and our attorney's offices" (EFF). EFF's account is its own, and Axon's September 2025 post describes an "unalterable digital audit trail" that records "who used the tool, when they used it, and what evidence was involved." Axon's product page, as archived on October 23, 2026, still says use is logged "without storing the draft itself." California's law requires the agency to keep the first draft, and we found no Axon page on November 13, 2026 that explains how a California agency does that with Draft One.
Can an officer's ChatGPT use come out in court?
It already has. In a November 20, 2025 opinion in Chicago Headline Club v. Noem, Judge Sara Ellis wrote in footnote 9 that "in at least one instance, an agent asked ChatGPT to compile a narrative for a report based off of a brief sentence about an encounter and several images," citing Axon body camera video. To the extent agents use ChatGPT for use of force reports, she wrote, "this further undermines their credibility and may explain the inaccuracy of these reports" (N.D. Ill. opinion, page 11 of 233). The prompt was captured on camera, and the court cited that video.
King County's memo names the cost for an officer. An AI error that an officer certifies leaves them facing "Brady/PID ramifications," with no draft to show "that theirs was an error of oversight, and not falsehood" (GeekWire).
The chat itself is a record somewhere. In federal prosecutions, the Jencks Act lets the defense obtain a government witness's "statement," defined to include "a written statement made by said witness and signed or otherwise adopted or approved by him," once that statement is in the government's possession (18 U.S.C. 3500). Whether an officer's chatbot prompt counts is a question for the prosecutor and the court. A personal ChatGPT account keeps the chat on OpenAI's servers until you delete it, then removes it "within 30 days" unless OpenAI must keep it, and OpenAI may share data with government authorities "if required to do so to comply with a legal obligation" (OpenAI, archived copy). How that legal process works is in can AI chats be subpoenaed.
Are police officers' AI chats public records?
In some states, yes, even on a personal account. Washington State Archives says a generative AI interaction is a public record if it "relates to the conduct of public business," and that work-related ones "created or received using a personal device/account are still public records" (Washington State Archives, updated November 2026). How other states and federal agencies treat AI records is in can government employees use ChatGPT.
Seattle shows how much of an officer's AI use can end up in public files. In 2024, a complaint alleged that a sergeant "used AI to draft emails and Blue Team reports." Seattle IT scanned the computers he'd used for visits to AI sites and found none on the eight it could reach, and the sergeant told investigators he used Grammarly, ChatGPT, and Google AI, drafting "without sensitive information included" and adding it after the output came back. The Office of Police Accountability found the allegation not sustained and recommended that the department write an AI policy. Its closed case summary, issued February 21, 2025 and posted on the city's website, describes all of it (Seattle OPA).
Can police officers use ChatGPT for personal questions?
Yes, on your own phone or computer, for questions with nothing from work in them. The AC-20 control above is about CJI. A department computer is different, and Seattle's investigators had city IT scan the sergeant's work computers for AI sites. What an employer can see on a work device is in can my employer see my ChatGPT.
A personal account still keeps your questions on the company's servers under your name. For an officer, those questions can be a pension rule, a custody schedule, a resume for a lateral move, or a health worry after a hard call.
How Morse keeps an officer's own questions out of stored chat histories
Every route above (a records request, a subpoena, a discovery motion) reaches a stored chat, and a work chat can be a public record whoever stores it, as Washington's guidance above shows. Morse is built so your chat history isn't stored on its servers at all.
Morse is for your own questions on your own device. Criminal justice information belongs only in the systems your agency approved under its CJIS rules, and a work chat can be a public record or discoverable wherever you type it. Morse makes no CJIS or other compliance claim, so no Morse mode is a place for CJI, case details, or anything from a report. For the rest of your life, you pick a privacy mode per chat:
- Private, for the pension, contract, or custody question you'd rather no AI company keep. Every new chat opens in Private, on open-weight models like DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing of what you ask. Where a provider confirms zero retention on each reply, Morse checks that confirmation and refuses a reply without it.
- E2EE, for what you'd tell no one: a health worry after a bad call, or a letter you're not ready to send. Your browser checks the enclave's hardware attestation before anything leaves your device, then encrypts the message so only that enclave can read it. Our servers pass along ciphertext they can't read, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the protection depends on the enclave hardware and attestation working as designed.
- Anonymous, for asking GPT, Claude, or Gemini a general question without the provider knowing it's you. Our proxy strips your account, email, and IP address and sends the message under Morse's own account. The provider may keep it for a limited time to check for abuse, under its own policy, so leave your name, your agency, and your badge number out of what you write.
In every mode, your chat history is stored in your browser, encrypted, and our servers don't keep your messages or log what you write. History locks when you sign out, and opening it takes two keys, one of which never leaves your device. Morse doesn't train AI on your content, sell your data, or show ads (privacy policy). Because we don't store chats, we usually have very little to give authorities. Clearing your browser or deleting your account removes the chats for good, and Morse can't recover them, so copy anything you need to keep. Use your own device.
The Free plan is $0 with 15 messages a day. Pro, at $19 a month, adds E2EE chats and unfiltered models, and Max is $49 a month. On your own device, send a first message from the Morse home page without an account, or start free with no card.
Common questions
Is Microsoft Copilot CJIS compliant?
Microsoft's CJIS page names Azure Government, Dynamics 365 US Government, and Office 365 GCC as the environments its state CJIS agreements cover, and lists Delaware, Louisiana, Ohio, South Dakota, and Wyoming as states without one. The page doesn't mention Copilot, so ask your agency whether its own tenant's AI features are in scope before any CJI goes near them.
What should an officer do if CJI already went into ChatGPT?
Tell your agency right away. CJIS Security Policy v6.1 tells agencies to make personnel report suspected incidents "immediately but not to exceed one (1) hour after discovery," and deleting the chat doesn't erase it at once: OpenAI's privacy policy says deleted conversations leave its systems within 30 days unless it has to keep them longer.
Who decides whether an AI tool can hold CJI?
Your agency, under your state's CJIS Systems Agency. The FBI's policy sets a floor and leaves agencies "the latitude to institute more stringent security requirements," so two departments in different states can give opposite answers about the same product.