Can I Put My Disclosure Into ChatGPT?

When you are charged with a criminal offence, the Crown gives you disclosure: police reports, witness statements, officers’ notes, photos, videos and sometimes hundreds of pages more. It is a lot to read, and it is tempting to upload it to ChatGPT, Claude or Gemini and ask for a summary or a list of weaknesses. Please do not. Here is what happens to that material, and why it can create problems for your case, for the people named in it, and for you.

Your Disclosure Is Not Ordinary Paperwork

Disclosure exists so you can make full answer and defence. Under R. v. Stinchcombe, [1991] 3 S.C.R. 326, the Crown must give you all relevant information in its possession, whether it helps or hurts its case. That makes disclosure some of the most sensitive material you will ever hold: it often contains the full names, addresses, phone numbers, dates of birth, medical details and private statements of complainants, witnesses and police officers.

The law places limits on what can be done with some of it:

• Publication bans. In sexual offence cases, and in many others on request, the court can order under section 486.4 or 486.5 of the Criminal Code that information identifying a complainant or witness “shall not be published in any document or broadcast or transmitted in any way.” Bail hearings are often subject to a ban under section 517. Breaching these orders is an offence.
• Records of complainants. If records relating to a complainant are produced to the defence under sections 278.1 to 278.91, the judge can order that the accused “not disclose the contents of the record to any other person” without the court’s approval.
• Sensitive recordings. Ontario’s Crown Prosecution Manual directs that video or audio of child abuse or a sexual offence “must not be provided to an unrepresented accused without a court order,” with conditions such as “no copying or disseminating the material.”
• Disobeying a court order is itself an offence under section 127 of the Criminal Code.

No Canadian court has yet decided whether typing protected information into a chatbot counts as “transmitting” it under a publication ban. We are not suggesting it is automatically a crime. But it is a risk you do not need to take, and if you are under an order not to share a record with “any other person”, uploading it to a company’s servers is very hard to defend.

What Happens to What You Type

Consumer AI services are run by companies with their own rules about your data. Those rules change, and they are often not what people assume.

It May Be Used to Train Future Models

• ChatGPT (Free, Plus and Pro): OpenAI says that “we may use your content to train our models,” and that users can switch this off in their data controls. Its business, enterprise and API products are not used for training by default (OpenAI).
• Claude (Free, Pro and Max): in August 2025 Anthropic changed its consumer terms so chats can be used for model training if users allow it, and extended data retention “to five years, if you allow us to use your data for model training.” Its commercial and API products are excluded (Anthropic).
• Gemini apps: Google says chats may be used to improve its services, that “a subset of chats are reviewed by human reviewers,” that reviewed chats “are retained for up to three years,” and warns users: “Please don’t enter confidential information that you wouldn’t want a reviewer to see” (Google).

In other words, depending on the service and your settings, the details of an allegation, and the names of the people in it, may be kept for years, read by staff, and used to train models that millions of other people will query.

In May 2026, Canada’s Privacy Commissioner and three provincial regulators found that the way OpenAI initially trained ChatGPT did not comply with Canadian privacy laws, citing problems with consent, transparency, accuracy and overcollection of personal information (Office of the Privacy Commissioner).

It Is Not Privileged

Your conversations with your lawyer are protected by solicitor-client privilege. That protection requires a confidential communication between a lawyer and client for the purpose of legal advice (R. v. Fox, 2026 SCC 4). A chatbot is not your lawyer.

OpenAI’s chief executive acknowledged this publicly in 2025, saying that there is legal confidentiality for conversations with doctors and lawyers, “and we haven’t figured that out yet for when you talk to ChatGPT” (TechCrunch). In February 2026, a U.S. federal judge ruled that a fraud defendant’s conversations with an AI chatbot were not protected by privilege, reportedly noting that “no such relationship exists, or could exist, between an AI user and a platform” (Bloomberg Law). That is American law, and no Canadian court has decided the point, but the Canadian test for privilege points the same way.

It Can End Up in Court, and With Police

AI companies respond to legal demands for user data like any other company. OpenAI’s own transparency report shows that in the second half of 2025 it received 75 requests from authorities for the content of users’ accounts and disclosed data in 62 of them (OpenAI).

Prosecutors in the United States have already relied on suspects’ chatbot conversations. In the federal case over the 2025 Palisades fire in Los Angeles, the complaint cites the suspect’s ChatGPT prompts, and OpenAI said it “responded to standard law enforcement requests related to this individual” (Axios). Chats can also be found on a seized phone or laptop, whichever company stored them.

Companies also review conversations themselves, and some will go to police without being asked. OpenAI has said that where reviewers determine a conversation “involves an imminent threat of serious physical harm to others, we may refer it to law enforcement” (OpenAI). Anthropic’s privacy policy, effective September 10, 2026, says it may share personal data with law enforcement where it believes in good faith that disclosure is reasonably necessary to “prevent serious harm to any person or to property,” and its law-enforcement policy makes an exception to its usual requirement of a subpoena or warrant where “there is an emergency that may result in imminent physical harm or death” (Anthropic; Claude Help Center).

This is no longer hypothetical. In late September 2026, a woman in Bonita Springs, Florida, who reportedly used Claude as a kind of diary, was arrested after an entry allegedly saying she would “shoot up” the Lee County Sheriff’s Office. According to news reports citing the arrest report, Anthropic’s systems flagged the conversation, a human reviewer assessed it, and the company alerted authorities; deputies then detained her at her home and she was charged with making a written threat (TechRepublic; The Next Web). The allegations have not been proven, and Anthropic has not commented publicly on the case. The Lee County Sheriff’s comment was blunt: “Users need to understand that you are never truly anonymous.”

Most people asking about their own charges will never trigger a referral. But people facing charges are often angry, frightened and venting, sometimes about the complainant or the police. A message written in that state, read by a stranger out of context, is exactly the kind of thing you do not want stored on a company’s servers.

It Can Leak, or Be Repeated

Stored data does not always stay where it is supposed to:

• In March 2023, a ChatGPT bug let some users see other users’ chat titles (OpenAI).
• In 2025, ChatGPT conversations that users had shared with a “discoverable” option appeared in Google search results until OpenAI withdrew the feature (TechRadar), and hundreds of thousands of shared Grok conversations were indexed the same way (Fortune).
• AI tools have also attached false criminal allegations to real people. A Norwegian man complained to regulators after ChatGPT falsely said he had murdered two of his children (noyb), and a defamation suit against Google over AI-generated claims was allowed to proceed in 2026 (Bloomberg Law).

Feeding a model the allegations against you, and the names of everyone involved, adds that information to systems with a track record of mixing up and repeating what they have seen.

”But AI Is Good at Summarizing Documents”

It is. But a summary is not the same as a defence review. What matters in disclosure is often what is missing: the officer’s notes that do not mention a key detail, the timing that does not line up, the witness who was never interviewed, the video that was requested but not produced. Finding those gaps depends on knowing what should be there, how the local police service usually documents a case, and what the Crown is likely to do with it. Our article Can ChatGPT Defend My Criminal Charges? explains why that knowledge is not something an AI model has learned.

What to Do Instead

1. Give your disclosure to your lawyer. Reviewing it is part of what they do, and your discussions about it are privileged. See our guide to Crown disclosure.
2. Keep it private. Do not post it, share it with friends or family, or upload it to any app or website.
3. Write your own notes on paper or offline, marked as prepared for your lawyer, and give them only to your lawyer. Our article on collecting evidence for your lawyer explains how.
4. If you use AI at all, keep it general. Asking “what does a peace bond mean?” is very different from pasting in a witness statement.
5. If you have already uploaded something, delete the conversation, turn off training in the app’s settings, and tell your lawyer what you shared. Deletion may not remove copies that have already been kept or reviewed, which is why your lawyer needs to know.

How Mor Fisher Can Help

We review disclosure in detail on every file, and we know what local police services and Crown offices normally produce, which is what makes a gap stand out. If you are representing yourself and are unsure what to do with your disclosure, a free consultation can help you understand what you have and what to ask for next.

Your disclosure belongs with your lawyer, not a chatbot. Contact Mor Fisher LLP at 705-721-6642 or through our contact page to have it reviewed in a conversation that stays privileged.