ChatGPT legal: what OpenAI now offers law firms, what lawyers use it for, and what went wrong
Lawyers who use ChatGPT ask three things: what OpenAI now sells to law firms, what the tool is actually good for, and what happened to the lawyers who were sanctioned. Here are all three, each with a date and a source, and the one thing ChatGPT does not keep for you: what each client sent you.
This page belongs to our AI for lawyers guide, which maps the tools and the bar guidance. Here the subject is one product and the news around it: OpenAI's law-firm offering, the uses lawyers describe, the court cases, and the settings that change what OpenAI keeps. If you want prompts for day-to-day client work, those are in ChatGPT for lawyers; this page has none.
Two things to know up front. Nothing here is legal advice, and this page draws no conclusion about whether any use of ChatGPT is permitted: that is for your jurisdiction's rules, your court and your firm. And it is published by Equerry, a personal assistant that is not a legal research or drafting tool. Each OpenAI detail below is quoted from OpenAI's own pages as they stood on 17 October 2026; other dates are marked.
What OpenAI offers law firms today
OpenAI's page for law firms, the first result for "chatgpt legal" when we checked on 17 October 2026, opens with "Bring frontier intelligence to your law firm" and promises "controls designed for your professional obligations." The product behind it is Astra for Law, announced on 17 September 2026 (LawNext covered the launch the same day). OpenAI's announcement describes it as combining "GPT‑6 Astra, our latest and most powerful model, with settings, tools, and context tailored for professional legal work." LawNext, reporting from OpenAI's media briefing, wrote that Jason Boehmig, the executive who leads OpenAI's legal products (hired in June "to lead development of products for the legal industry"), "emphasized that this is not a new model, but rather a model configuration."
| What OpenAI lists | What its pages say (read 17 October 2026) |
|---|---|
| Legal search index | Astra for Law "can search U.S. case law, statutes, regulations, court rules, and administrative decisions across a corpus of more than 230 million URLs, with sources added daily," with case law from Free Law Project, the nonprofit behind CourtListener. |
| Who can use it | "Initially available to selected law firms through Trusted Access in ChatGPT and Codex," with API access to follow, OpenAI says. The law page: "Eligible lawyers and people working under their supervision at eligible law firms." LawNext reported the program "is targeted at Am Law 200" firms. |
| Data handling | The law page: "API usage under our legal offering comes with zero data retention," and "ChatGPT Enterprise usage comes with an 'eyes-off' human review posture by default." Asked about exceptions, the executive who leads OpenAI's legal products told reporters the zero-retention line summarises an access agreement of some 30 pages, and, LawNext reports, said OpenAI "has worked with the industry to validate that it works in this context." "It's definitely more complicated than the one sentence," he said, "but we also feel extremely confident that the 30 pages do meet the need" (LawNext, 17 September 2026). |
| Plugins and partners | 26 partner-built plugins, plus nine community plugins with 47 custom skills. The law page names Clio, DeepJudge, Free Law Project, iManage, Intapp, Laurel, LegalQuants, Legora, Litera, Relativity and Thomson Reuters. |
| ChatGPT for Word | Made "generally available today" in the same announcement, "so lawyers can proofread, get suggested edits, and flag formatting issues." |
| Firm builds | OpenAI describes tools built with Sullivan & Cromwell (an agreement analyzer), Ropes & Gray (deal diligence) and Cooley (IPO drafting), and governance work with Latham & Watkins. |
OpenAI also published its own test. On 200 legal research questions from Vals AI's Legal Research Bench, "Astra for Law passed the evaluation's overall correctness check on 54.0% of questions, compared with 38.7% for GPT‑6 Astra using web search alone." That is OpenAI's measurement of its own product. On that test, the answer did not pass the correctness check on 46% of questions; read every authority yourself.
What this means for a solo or small-firm lawyer: the pages describe access for "eligible law firms," not a plan you can buy at checkout. Neither page mentions Plus or any personal plan; Astra for Law is offered through the Trusted Access program. The comparable move by Anthropic is covered in Claude for Legal.
What lawyers use ChatGPT for, and where it fails
The third result for "chatgpt legal" is "Legal ChatGPT: Tips, Prompts, and Use Cases," an article by the practice-management company MyCase, published on 21 March 2025 in the ABA Law Practice Division's Law Technology Today. It names six common uses: "Creating, editing, and summarizing legal documents," "Performing legal research and discovery," "Transcribing voice and video recordings," "Assisting with client communications," "Analyzing case details and predicting outcomes" and "Drafting marketing content." It also warns that chatbots "are prone to hallucinations, have trouble understanding contextual nuances of the law, and have trained-in biases," and among its eight prompt-writing steps lists "Scrub prompt for confidential information," "Request citations and dig into the sources" and "Evaluate and adjust the outputs before using."
That list mixes two kinds of work, and the difference decides how much checking each one needs:
- Work you can check by reading it. A first draft built from your own outline, a summary of a document you pasted in full, a list of questions for a first client meeting, a plain-English version of a process you already know. You hold the source, so every error is visible to you.
- Work you can only check by redoing it. Research, case analysis and anything presented as "the law says." The output reads the same whether it is right or wrong, and nothing in a general chat checks it against a court's record. This is where the sanction cases below come from.
Why lawyers were sanctioned: the cases as reported
New York, June 2023. Reuters reported on 22 June 2023 that U.S. District Judge P. Kevin Castel in Manhattan imposed sanctions on two New York lawyers and their firm over "a legal brief that included six fictitious case citations generated by an artificial intelligence chatbot, ChatGPT." The fine was $5,000 in total. The judge found they acted in bad faith and made "acts of conscious avoidance and false and misleading statements to the court"; the firm said it "respectfully" disagreed that they had acted in bad faith. The same order, Reuters reported, said there is nothing "inherently improper" in lawyers using AI "for assistance," and required the lawyers to notify the real judges named as authors of the fake opinions.
California, September 2025. CalMatters reported that California's 2nd District Court of Appeal fined a Los Angeles-area attorney $10,000, a fine it said "appears to be the largest issued over AI fabrications by a California court," after finding that "21 of 23 quotes from cases cited in the attorney's opening brief were made up." The court wrote: "We therefore publish this opinion as a warning," and that no filing "should contain any citations— whether provided by generative AI or any other source—that the attorney responsible for submitting the pleading has not personally read and verified." The attorney told CalMatters he wrote the appeal and then used ChatGPT to try to improve it, and did not know it would add case citations.
Still happening. Damien Charlotin's AI Hallucination Cases database, which CalMatters cited, keeps adding court decisions about fabricated or misrepresented citations. Its newest entries, read on 17 October 2026, are dated as late as 15 October 2026, and they include filings by lawyers as well as by people representing themselves.
The fact the reports share: in each, citations went to a court without anyone opening the cases they named.
Is ChatGPT legal for lawyers? What the sources say, and who decides
The fourth result for "chatgpt legal" asks exactly this. OpenAI's own usage policies (last changed 29 October 2025) prohibit using its services for the "provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional." Spellbook's guide (updated 8 October 2026) sets out three separate authorities a lawyer answers to on the question, the vendor's usage policy, "the ethics rules of the lawyer's own jurisdiction, and the standing order of the judge assigned to the matter," and adds that "satisfying one of the three is not satisfying the other two." Whether your use is permitted is decided by your jurisdiction's rules and bar guidance, your court and your firm's policy; this page does not decide it.
Before you paste: one test, and the controls OpenAI publishes
One question before every paste does most of the work: would this tell a stranger who the client is, what the matter is, or what you intend to do? If it would, it stays out of a personal ChatGPT account unless your firm has approved a specific tool and plan for it. A protective order or a nondisclosure agreement sets its own terms for the documents it covers, and those terms, not a chat setting, decide where the documents may go. OpenAI's controls change what OpenAI keeps; they do not change what you are allowed to share.
Started as Temporary and set to Unpersonalized before the first message, a chat stays out of your history and memory for as long as it remains temporary, though OpenAI's Data controls article says such a chat "may be retained for up to 30 days for safety purposes" (our Temporary Chat guide has the steps). Turning off Improve the model for everyone in Data controls keeps your new conversations out of model training without deleting or hiding saved chats; how that switch, feedback and business plans fit together is set out line by line in what OpenAI publishes about ChatGPT and confidentiality. For firm workspaces, OpenAI says: "By default, OpenAI does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models," and "Your organization's privacy, retention, and access policies still apply."
The client record: what each client sent you, and where it lives
ChatGPT remembers. OpenAI's Memory article says it can draw on past chats, saved memories, files and "content from connected apps, such as Gmail," and that "Memory does not retain every detail from every conversation." OpenAI's law offering reaches further for large firms, through plugins into document systems and matter files. The question for a solo or small-firm lawyer is control. To keep up with a client through ChatGPT, you either paste or upload each message by hand, or you connect a whole mailbox, which is an account-wide decision (we cover what that connection grants in can ChatGPT read my emails).
Before a call, the record you want is narrower than a mailbox and steadier than a pasted chat: what the client said about the closing date, whether the signed lease came back, which documents they sent last week, what you promised by Friday. That record is about the client, not the law, and none of the research tools on our AI for lawyers map is built to hold it.
Equerry is a personal assistant who helps you get things done. In a practice, the part it takes on is that client record, built only from what you choose to send it:
- One thread at a time. CC your own Equerry address (it ends in @mail.getequerry.app) on a live client thread, or forward a thread after the fact. Replies that keep the address on the thread keep arriving. It never asks for your inbox, and the original stays in your mailbox. What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line.
- A voice note after the call. Record a short voice note in Equerry on your way out: what was agreed, what the client is worried about, what you owe. Equerry does not join or record your calls.
- Documents and notes. Share a PDF or a screenshot from your iPhone's share sheet, or type a note. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original.
- A brief before the meeting, if you want one. Meeting Prep is a Routine you set up yourself; with your permission to read your calendar, it sends a short brief before a meeting, built only from what you sent.
| What you send | The client's thread about the lease renewal, with your Equerry address on the CC line; the marked-up draft they attached, shared as a PDF from your iPhone; the voice note you recorded in Equerry after Tuesday's call; the note you typed after the first meeting. |
|---|---|
| What you ask | "Before I call the client back: what did they say about the renewal date, has the signed lease come back, and what did I say I would send?" |
| What you get back | The renewal date the client gave, taken from the CC'd thread; whether the signed lease has come back from the landlord's side on that thread, if their reply kept your Equerry address on it; and what you said you would send, from the note you typed after the first meeting. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original. For the thread: What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line. |
The limits, plainly. Equerry is not a legal research or drafting tool: it finds no cases and writes no briefs. It has no connection to a document management system, to Clio, to Outlook or to any CRM, and it never sends, books or files anything for you. It knows only what you sent it, and the other people on a CC'd thread will see its address. What you send is processed to answer you, including by the outside providers named on our AI disclosure page. Privilege, confidentiality duties and bar rules are not questions Equerry answers, and it claims nothing about them. Sending client material to it is a call you make under your firm's policy, exactly as with ChatGPT; the difference is how much that call covers: a thread or a note you picked, not everything an account can reach.
Where your firm's policy allows it, use ChatGPT, or OpenAI's law offering if your firm has it, for drafting you can check line by line, and read every case it names before it goes anywhere. Keep the client record (what each person sent, said and was promised) in Equerry. It is on the App Store for iPhone and iPad.
Sources, all read 17 October 2026 unless stated. OpenAI: OpenAI for law firms (Trusted Access, data handling, partners; no publication date shown); Introducing Astra for Law (product, legal search index, benchmark, plugins, ChatGPT for Word, firm builds); Usage policies (changelog entry 2025-10-29). OpenAI Help Center: Temporary chat in ChatGPT (8914046); Data controls in ChatGPT (7730893); Memory in ChatGPT (8590148). LawNext, 17 September 2026. MyCase, "Legal ChatGPT: Tips, Prompts, and Use Cases", Law Technology Today, ABA Law Practice Division, 21 March 2025 (read in the Internet Archive's copy of 15 August 2026). Spellbook, "Is Using ChatGPT Legal for Lawyers in 2026?", updated 8 October 2026. Sara Merken, Reuters, "New York lawyers sanctioned for using fake ChatGPT cases in legal brief", 22 June 2023 (read in syndication). Khari Johnson, CalMatters, "California issues historic fine over lawyer's ChatGPT fabrications", 22 September 2025, and the 2nd District opinion, B331918. Damien Charlotin, AI Hallucination Cases database. OpenAI changes its products and pages often; where its current page disagrees with this one, OpenAI's page is right. This page is general information, not legal advice.
Frequently asked questions
Is it legal for lawyers to use ChatGPT?
No vendor and no web page can settle that for your practice, and this one does not try. Here is what the published sources say. OpenAI's usage policies (last changed 29 October 2025) bar using its services for the "provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional." Spellbook's guide "Is Using ChatGPT Legal for Lawyers in 2026?" (updated 8 October 2026) treats the question as three separate sets of conditions: the vendor's usage policy, the ethics rules of the lawyer's own jurisdiction, and the standing order of the judge assigned to the matter. Whether a particular use is permitted is decided by your jurisdiction's rules and bar guidance, the court you file in and your firm's policy. This is general information, not legal advice.
What does OpenAI offer law firms?
As read on 17 October 2026: Astra for Law, announced on 17 September 2026, which OpenAI describes as combining "GPT-6 Astra, our latest and most powerful model, with settings, tools, and context tailored for professional legal work," including a legal search index over U.S. case law, statutes, regulations, court rules and administrative decisions. It is "initially available to selected law firms through Trusted Access in ChatGPT and Codex," with API access to follow. Alongside it OpenAI lists 26 partner-built plugins (its law page names Clio, iManage, Relativity and Thomson Reuters among the partners), nine community plugins, and ChatGPT for Word, which it made generally available the same day. Eligibility is for "eligible law firms"; LawNext reported that the program targets Am Law 200 firms.
What happened in the ChatGPT sanction cases?
In the two cases reported most widely, the lawyers filed text that cited cases nobody had checked. Reuters reported on 22 June 2023 that a federal judge in Manhattan fined two New York lawyers and their firm $5,000 in total over a brief with six fictitious case citations generated by ChatGPT. CalMatters reported in September 2025 that California's 2nd District Court of Appeal fined a Los Angeles-area attorney $10,000 after finding that 21 of 23 quotations from cases cited in his opening brief were made up; he told the court he had not read the text ChatGPT produced before filing. New decisions on fabricated citations are still being added to Damien Charlotin's public database, with entries dated as late as 15 October 2026 when it was read for this page.
What do lawyers keep out of ChatGPT?
Whatever their firm's policy names; no general list outranks it. For everything the policy does not cover, one screen helps: could a stranger reading this work out the client, the matter or your next move? Material that could usually stays out unless the firm has signed off on a particular tool and plan. In practice that means a client's identity next to the facts, advice and other privileged exchanges, negotiating positions, and numbers that pin down a person. Documents covered by a court order or a nondisclosure agreement come with their own conditions, and those conditions decide whether any tool may see them. Public law you pulled up yourself, and your own drafts with placeholders where the names were, raise fewer questions.
Does ChatGPT keep what I paste into it?
By default, a regular chat stays in your history until you delete it. OpenAI's Data controls article (read 17 October 2026) says that with "Improve the model for everyone" turned off, "your new conversations won't be used to train OpenAI models. They can still appear in your chat history," and a thumbs up or down can still send "the entire conversation associated with that feedback" to training. A temporary chat stays out of history and memory while it remains temporary, but "may be retained for up to 30 days for safety purposes." On Business, Enterprise, Edu and ChatGPT for Healthcare workspaces, OpenAI says it does not use content to train its models by default. Our guide to whether ChatGPT is confidential reads OpenAI's data terms in full.
Does ChatGPT remember what each client said?
It can remember things, but it does not keep a record organised by client. OpenAI's Memory article (read 17 October 2026) says ChatGPT can draw on past chats, saved memories, custom instructions, files in Library and "content from connected apps, such as Gmail," depending on plan and region, and that "Memory does not retain every detail from every conversation." What it remembers comes from what you typed, uploaded or connected. A personal assistant like Equerry works the other way round: it keeps only the threads, notes and documents you choose to send it, and answers from them when you ask.
Last updated 2026-10-17.
All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.