OpenAI's usage policy update: what changes for ChatGPT users on November 20, 2026
The Morse teamPublished 10 min read
The OpenAI usage policy update takes effect on November 20, 2026, replacing the version in force since October 29, 2025. Five changes reach an ordinary ChatGPT user: a spelled-out ban on using ChatGPT's answers to train rival models, including by "providing feedback used in training"; a new line against trying "to extract protected reasoning"; a licensed-advice rule rewritten to target people "building or offering AI tools" for others; a child-safety section split in two; and a wider definition of high-stakes automation. The four opening paragraphs, including the one on how OpenAI monitors and enforces its rules, are the same word for word.
On November 11, 2026, we read the live policy, the October 29, 2025 text that OpenAI still hosts at its own dated address, the help center's summary of the update, and OpenAI's pages on enforcement, account actions, and data retention, and compared the two versions line by line. Wayback Machine copies of OpenAI's page show the old text at 15:44 UTC on October 21, 2026 and the new text by 22:55 UTC, which matches the changelog's date. OpenAI gave good notice. It posted the new text 30 days before it applies, kept the old one a click away under "Previous Version," and wrote a changelog entry that names each change.
What changed in OpenAI's usage policies?
Mostly rules for developers, platforms, and governments, plus a handful of lines that reach anyone with a ChatGPT account. The changelog entry, dated October 21, 2026, says the update clarifies "restrictions on using OpenAI models and outputs to train other models," "protections for children and teens," and "requirements for high-stakes uses," and adds approval rules for "national security and law enforcement uses" and "model-access intermediaries." The table sets the October 29, 2025 version beside the November 20, 2026 version, section by section, both read on November 11, 2026.
| Section | October 29, 2025 wording | November 20, 2026 wording |
|---|---|---|
| Opening principles | "We monitor and enforce policies with privacy safeguards in place and clear review processes" | Unchanged, along with the lines on appeals and on losing access |
| Licensed advice | Bans "provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional" | Bans "building or offering AI tools that provide others with tailored advice that requires a license," and adds "You also may not state or imply that the AI itself holds professional credentials or authorization" |
| Government use | Bans "national security or intelligence purposes without our review and approval" | Bans "national security, law enforcement, or intelligence purposes without our review and approval" |
| Children | One section, "Keep minors safe," saying the services "must never be used to exploit, endanger, or sexualize anyone under 18 years old" | Retitled "Keep young people safe" and split into "Child sexual abuse and exploitation" and "Teens' use of our products and services" |
| Reports to NCMEC | "We report apparent child sexual abuse material and child endangerment" | "We report apparent child sexual abuse material and other child sexual abuse conduct" |
| Child abuse list | CSAM and "grooming of minors" | Adds "sextortion or any form of conduct based child sexual abuse," "sexualizing children," and "other usage or behaviour that supports the sexual abuse or exploitation of children" |
| Teens | Separate bans on "dangerous challenges for minors" and "underaged access to age-restricted goods or activities" | One merged ban, under a new line: "We build age-appropriate safeguards designed to reduce exposure to content that may be harmful or developmentally inappropriate" |
| High-stakes uses | "automation of high-stakes decisions in sensitive areas without human review," with law enforcement as its own area | "high-stakes actions," covering "actions, decisions, transactions, submissions, and other steps that materially affect a person's rights, safety, access to essential services, or other consequential interests." Law enforcement joins "national security, intelligence, and law enforcement" |
| Political campaigning | Banned, with no detail | Same words, now linked to a help page of examples |
| Training other models | Absent from the usage policies. The Terms of Use bar using output "to develop models that compete with OpenAI" | A new section on outputs used to "generate, prepare, select, or supply training material or signals" for models meant as "alternatives to or substitutes for OpenAI models or services" |
| Hidden reasoning | Nothing | "You must not attempt to extract protected reasoning through unauthorized access, reverse engineering, or circumvention of safeguards" |
| Exceptions | Nothing | "OpenAI may grant specific partners case-by-case exceptions," with national security uses held to OpenAI's principles against "mass domestic surveillance" |
| Model resellers | Nothing | A "model-access intermediary platform" needs "our prior review and approval" |
| Privacy of others | Bans facial recognition databases without consent, real-time remote biometric identification in public, and social scoring | Unchanged |
OpenAI's help center summary calls the training rules "existing restrictions" that it "clarified," and says "the previous version remains in effect" until November 20.
Does the update change how OpenAI monitors your chats?
No. The policy's one sentence on monitoring, "We monitor and enforce policies with privacy safeguards in place and clear review processes," reads the same in both versions. The detail sits on other pages. OpenAI's transparency page, last updated October 27, 2026, lists "classifiers, reasoning models, hash-matching, blocklists, and other automated systems," and says "our team may review flagged content." A help article updated November 10, 2026 adds that "ChatGPT can take certain safety actions automatically, without a person reviewing the conversation first" (OpenAI).
The update changes the list those systems check against, and a ban under that list can keep a chat on OpenAI's servers. OpenAI's privacy policy, updated September 10, 2026, says that "if specific Content, or your account, is banned because of violations of our usage policies," OpenAI may keep that data after you delete it. From November 20, a chat banned under a new rule, such as the ones on training data or protected reasoning, falls under that clause like any other. How long seven providers keep flagged content is in what is AI abuse monitoring.
Can you still ask ChatGPT medical or legal questions?
Yes, and the new wording says so more plainly. The October 2025 line banned "provision of tailored advice that requires a license, such as legal or medical advice," and within days, posts claimed ChatGPT had stopped answering medical and legal questions. On November 3, 2025, OpenAI's head of health AI, Karan Singhal, replied "Not true," adding "Model behavior remains unchanged" and "ChatGPT has never been a substitute for professional advice."
The November 2026 text aims the rule at whoever is "building or offering AI tools that provide others with tailored advice that requires a license," and drops the words "such as legal or medical advice." It also bars claiming "that the AI itself holds professional credentials," so a chatbot that presents itself as a licensed therapist would break it. Asking about your own symptoms, lease, or tax form doesn't build a tool for anyone. This isn't legal advice, and what to leave out of a legal question is in private AI for legal questions.
Can you get banned for training another AI on ChatGPT's answers?
Yes, if the model is meant to replace OpenAI's, and that was true before the update. The Terms of Use, effective January 1, 2026, bar using output "to develop models that compete with OpenAI," and let OpenAI "suspend or terminate your access" if you breach the terms "or our Usage Policies."
The new section spells out how far the rule reaches. It covers "providing feedback used in training, including labels, scores, rankings, preferences, rewards, or corrections." It applies "even if you do not train the resulting model," and when "outputs from product use or evaluation are later used or supplied for those training purposes." Read plainly, a freelancer doing paid labeling work for another AI lab who has ChatGPT write the rankings is covered, and so is someone who saves ChatGPT answers and later sells them to a rival as training data. The policy still allows "research, software development, model evaluation and benchmarking, and product development, within these boundaries."
Is it against the rules to ask ChatGPT to show its reasoning?
Only if you try to get around the block. The new line bans attempts "to extract protected reasoning through unauthorized access, reverse engineering, or circumvention of safeguards." Both versions also ban "circumventing our safeguards." OpenAI's account help page, updated November 10, 2026, lists "Attempts to bypass safety filters, content restrictions, or other safeguards" among the reasons for a warning, and says "Continued violations after a warning may lead to further restrictions or permanent account closure."
What changed for teens and children?
The section was split and its reporting line reworded. Child sexual abuse now has its own subsection, naming "sextortion" and "sexualizing children," and the teen subsection opens with "OpenAI tools must not be used to endanger anyone under 18." The sentence on reports to the National Center for Missing and Exploited Children (NCMEC) changed from "child sexual abuse material and child endangerment" to "child sexual abuse material and other child sexual abuse conduct." Neither version says what a report contains. OpenAI's report counts, and when it goes to police, are in does ChatGPT report you to the police.
The new "age-appropriate safeguards" line describes systems already running. As of November 11, 2026, ChatGPT's age prediction can move an adult's account into ChatGPT for Teens until they verify with a selfie or an ID through Persona; we compare it with other apps in AI age verification and privacy. For a teen account linked to a parent, the parent "may receive a safety notification" when ChatGPT ends a chat or suspends the account over violence, but it "does not give your parent or guardian access to your conversations, transcripts, or chat history" (OpenAI).
What happens to your account if OpenAI thinks you broke the rules?
For some violations, a warning comes first, then a suspension or a closure. The account help page says "A warning does not, by itself, mean your account has been suspended or closed," and "Discussing a sensitive topic does not, by itself, mean your account will be permanently closed."
For violence, ChatGPT now acts on its own. It "shows a warning before ending a conversation," and after "repeated conversation endings" it may suspend the account for 24 hours and sign you out at once. An ended chat "no longer appears in your chat history or search results, and you cannot reopen or continue it." As of November 11, 2026, the help page says this is "gradually rolling out" (OpenAI).
To appeal, use the link in the notice email, or OpenAI's appeal form if you can't reach the email. The help page warns "Requesting a review does not guarantee that your account will be restored." The Terms of Use also let OpenAI end access if "Your use of our Services could cause risk or harm to OpenAI, our users, or anyone else," a ground separate from breaking the terms or the usage policies.
Does the update let police use ChatGPT?
Only with OpenAI's approval. Law enforcement joins national security and intelligence among the purposes that require "our review and approval," and approved uses follow OpenAI's Principles for National Security Partnerships, which say OpenAI "will not support use of OpenAI tools for" "mass domestic surveillance."
Those rules govern agencies as customers. Whether OpenAI hands your own data to police is set by the privacy policy, which this update leaves alone. It lets OpenAI share data with "government authorities" if, among other reasons, "we determine, in our sole discretion, that there is a violation of our terms, policies, or the law." That covers the new rules as well. Warrants and subpoenas for chats are in can AI chats be subpoenaed, and the privacy policy's own changes are in OpenAI privacy policy changes.
How Morse keeps your chat history on your device
OpenAI enforces its usage policies on the chats it keeps. Its classifiers check them against these rules, its team may review the flagged ones, and a chat banned under the policy can stay on OpenAI's servers after you delete it. With Morse, your chat history lives in your browser, encrypted, and our servers don't keep messages or log their content. The account holds sign-in details, your plan, your balance, and a usage ledger with no content (privacy policy). Morse doesn't train AI on your chats, sell your data, or show ads, and because we don't store chats, we usually have very little to give authorities.
Each chat runs in the mode you pick:
- Anonymous, to use GPT, Claude, or Gemini without your name on the words. Our proxy strips your account, email, and IP address and sends the message under Morse's own account. The provider sees what you wrote without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy. The words themselves still reach the provider, so leave out names and details you'd rather keep to yourself.
- Private, the default, for chats no model host should keep. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, answer in this mode. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
- E2EE, for questions no reviewer should read. Before anything leaves your device, your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The guarantee rests on the enclave hardware and attestation working as designed. E2EE chats come with Pro, at $19 a month.
History locks when you sign out, and opening it takes two keys, one of which never leaves your device. Clearing your browser or deleting your account deletes your chats, and Morse can't recover them, so keep anything you need. Like OpenAI, Morse reports CSAM to NCMEC.
You can try Morse from its home page before making an account, or sign up free with no card.
Common questions
Do I have to accept OpenAI's new usage policies?
There's no way to stay on the old version. OpenAI's Terms of Use require you to follow the Usage Policies and say that if you don't agree to changes, "you must stop using our Services." Its help center says the October 2025 version applies until November 20, 2026.
Do the usage policies apply to ChatGPT Business and Enterprise accounts?
Yes. OpenAI's changelog says the October 2025 version created "a universal set of policies across OpenAI products and services," and the new version keeps that. The automatic 24-hour suspensions over violence don't apply in Business, Enterprise, Edu, or Healthcare workspaces, according to OpenAI's help center.
Can I use ChatGPT to write texts for a local election campaign?
Not at scale. OpenAI's political campaigning help page says the rules "apply to all users, not only official campaigns," and bans creating or automatically sending campaign emails, texts, and direct messages, "even if a human reviews them before they are sent." It allows drafting a speech or talking points that a person will deliver.
How can I find out about the next usage policy change?
OpenAI's usage policies page links to a sign-up form for update notices and keeps a dated changelog going back to February 2021. The "Previous Version" link next to the effective date opens the text each version replaced.