The short answer
Not without preparation. With preparation, yes, but not everything. Three questions decide.
1. Does the document hold client information: names, details, sums, deal terms, the position in a case?
2. Does the client agree that an outside AI service will process the documents?
3. What do the service terms say about storing, training on and disclosing data, and is there a data processing agreement?
If the document holds personal data of Russian citizens, the requirements of 152-FZ, Russia’s personal data law, apply: storage in databases in Russia (Part 5 of Article 18) and notifying Roskomnadzor, the regulator, before a cross-border transfer (Part 3 of Article 12). The fine for a company that breaks the localization rule under Part 8 of Article 13.11 of the Code of Administrative Offenses is up to $72,000.
For an advocate, Article 8 of the Law on Advocacy also applies: secrecy covers any information linked to helping the client, and only the client can lift it. So get the client’s consent to AI before the first upload, not after.
You can upload without reservations a template with no names or sums, a question about a legal rule with no facts of the case, the text of a law or a published court decision. The answer is the same if the question is “can I upload a contract to a neural network” from another vendor: look at the terms, not the name.
What happens to an uploaded file
The file goes to the vendor’s servers and is processed under its terms. After that, what happens to the text depends on the plan and the account settings, not on the lawyer’s intentions.
What service terms usually cover, and what to read before you upload:
1. How long prompts, answers and files are kept, including after a chat is deleted.
2. Whether conversations are used to train the model, and how to switch that off.
3. Access to conversations by the vendor’s staff and contractors.
4. Disclosure of data at the request of a court or a government body.
5. The country of the servers and a data processing agreement with your organization.
A free account and a business plan from the same vendor can process data under different rules. Check the terms of the plan the lawyer actually uses, not the general privacy page on the website.
A deleted chat does not mean deleted data: on 13 May 2025, a court ordered OpenAI to keep logs of ChatGPT output, including chats that users had deleted.
On 25 July 2025, OpenAI CEO Sam Altman said that conversations with ChatGPT have no legal protection like a conversation with a lawyer or a doctor, and that the company would have to disclose them if a court required it.
Three stories: Samsung, the lawsuit against OpenAI and the Heppner case
Samsung. Engineers uploaded internal source code to ChatGPT, and on 2 May 2023 the company banned generative AI for its staff. Once text is sent, you cannot take it back, and the ban came after the upload, not before it.
The New York Times lawsuit against OpenAI. The duty to keep output logs ended on 26 September 2025, but on 2 December 2025 the court upheld the disclosure of a sample of 20 million anonymized logs. Users’ conversations became material in someone else’s dispute.
United States v. Heppner. On 17 February 2026, Judge Jed Rakoff ruled that a defendant’s conversations with a public AI service were not protected by attorney-client privilege: the service’s policy allowed it to collect prompts and answers and disclose them to third parties.
All three stories are about processing terms, not about model quality. Read more about leak cases in our article about leaks through AI, and about privilege and secrecy in our review of attorney-client privilege and AI.
What to do instead
You do not need to give up AI: you need a tool whose terms fit professional secrecy and 152-FZ.
1. A model on the firm’s server or in a cloud in Russia: documents do not leave the network, there is no cross-border transfer, and the rule on databases in Russia is met.
2. Business access to a model with a data processing agreement, if the documents are anonymized and the client agrees.
3. Anonymization before sending: names, details and sums are replaced so that no one can link the text back to a person.
4. A written firm rule: which services are allowed, what must never be sent to them and who checks the result.
5. A log: who sent which document to the model and when, so that you can show the document’s path when a client or an inspector asks.
If your lawyers already use AI, the first step is simple: find out which services staff have open and which documents went there. After that, it is clear which processes to move to a closed model first.
How a closed model works and what it costs is on the page about a local neural network for lawyers. In any setup, the AI answer stays a draft: the lawyer gives the position and the signature.