Can you sue ChatGPT? What 12 AI apps' arbitration clauses say

The Morse teamPublished November 10, 202612 min read

You can sue ChatGPT's maker in small claims court, but as of November 2026, OpenAI's terms send almost every other claim to private arbitration and bar class actions, unless you opted out of arbitration within 30 days. The clause covers any claim "arising out of or relating to these Terms or our Services," and says "Class arbitrations, class actions, and representative actions are prohibited" (OpenAI). If you live in the EEA, Switzerland, or the UK, different terms apply, and they let you go to your local courts (OpenAI).

We read the US consumer terms of 12 AI chat apps on November 10, 2026, plus the European versions where an app keeps separate ones. For each we wrote down whether it forces arbitration and with which provider, whether it waives class actions, how and when you can opt out, whether small claims court is allowed, what happens when many people file at once, which law and court apply, and what changes for EU and UK consumers. Every cell in the table links to the terms it came from. This isn't legal advice.

If you want to sue over a ChatGPT answer, such as a false claim about you, these clauses decide where that claim can be heard and whether you can bring it with other users.

Can you sue OpenAI over ChatGPT?

Only in small claims court, unless you opted out. OpenAI's terms (effective January 1, 2026) send disputes to National Arbitration and Mediation, known as NAM, with one arbitrator, by videoconference if possible. Two kinds of claim stay out of it: "individual claims brought in small claims court" and requests to stop abuse of the service or intellectual property infringement (OpenAI).

For a privacy complaint, three more parts of OpenAI's clause come into play:

• It reaches back. It covers claims "regardless of when the claim arose, even if it was before these Terms existed."
• A court still decides some questions. The state or federal courts of San Francisco decide whether the class waiver is valid and hear any request for "public injunctive relief," after the arbitration of everything else.
• You must try to settle first. You send notice through OpenAI's informal dispute form, and either side can start arbitration only after 60 days.

OpenAI says it "will not seek attorneys' fees and costs in arbitration unless the arbitrator determines that your claim is frivolous." Clauses like this are enforceable in the US because the Federal Arbitration Act makes a written agreement to arbitrate "valid, irrevocable, and enforceable," except on the grounds that would undo any other contract.

Which AI apps make you give up your right to sue?

Seven of the 12 require arbitration for US users, and an eighth, xAI, bans class actions in court. Read on November 10, 2026, from each company's own consumer terms.

Anthropic, Google, and Meta, three of the largest companies here, keep you in court. Anthropic's consumer terms have no arbitration clause and no class-action waiver at all. They say disputes "will be resolved exclusively in the state or federal courts located in San Francisco, California" (Anthropic). Meta's terms let a consumer bring a claim "in any competent court" in their own country, and the new version Meta has posted for January 20, 2027 keeps that sentence (Meta). DeepSeek has no arbitration clause either, but its court is in China.

Can you opt out of ChatGPT's arbitration clause?

Yes, within 30 days of creating your account, by filling out OpenAI's opt-out form. As of November 10, 2026, the form asks for your first and last name, your account email, and an Organization ID, which it says is on your account's organization settings page on OpenAI's platform site. If you missed that window, you get another 30 days after any update to the arbitration terms takes effect. Opting out of an update leaves "the last set of agreed upon arbitration terms" in place (OpenAI).

The other apps that offer an opt-out each ask for it differently:

• Character.AI: email arbitration@character.ai within 30 days of first agreeing, "stating that you do not want to resolve disputes with Character.AI by arbitration" (Character.AI).
• Perplexity: email support@perplexity.ai or write to its San Francisco address within 30 days of registering. If you opt out of arbitration but not the class-action waiver, "the Class Action waiver still applies" (Perplexity).
• Mistral: email support@mistral.ai within 30 days with your name, address, and "a clear statement that you wish to opt out" (Mistral).
• Proton Lumo: a letter, not an email, to Proton AG, Attn: Legal, Route de la Galaise 32, 1228 Plan-les-Ouates, Switzerland, within 30 calendar days. Like Perplexity, Proton lets you opt out of the arbitration and the class waiver separately, and the waiver stays if you only drop the arbitration (Proton).
• Replika: email my@replika.com within 30 calendar days of accepting the terms, with your name, account email, and a clear statement that you want out (Replika).

Microsoft and xAI offer no opt-out when you sign up. Microsoft lets you reject a later change to its arbitration section "by personally signing and sending us notice within 30 days of the change by U.S. Mail" (Microsoft).

Is it too late to opt out of an AI app's arbitration clause?

For most of these apps, yes, unless you signed up in the last 30 days. Replika is the exception this month. Its terms dated October 28, 2026 let existing users opt out "within thirty (30) calendar days after these Terms of Service are first posted with this Section 12.9" (Replika). The version from March 30, 2026, as the Internet Archive captured it on May 1, had arbitration but no opt-out section. If the new terms went up on October 28, that window for existing Replika users closes around November 27, 2026.

We cover what else Replika's terms and privacy policy allow in is Replika private.

What happens when thousands of users file at once?

They wait in line. A class waiver means each user files alone, so a complaint shared by many users becomes many separate arbitrations. Four of the 12 have rules for that:

• OpenAI: if 25 or more people with the same or similar lawyers file similar claims within 90 days, NAM hears them in batches of up to 50, each "as a single consolidated arbitration with one arbitrator, one set of arbitration fees, and one hearing" (OpenAI).
• Microsoft: 25 or more "Related Cases" go in batches of up to 50, with each side picking 25. Nothing else can be filed until the first batch ends, then everyone mediates, with Microsoft paying the mediator (Microsoft).
• Character.AI: at 75 claimants or more, each side picks 10 test cases. "Until those cases are resolved, no other Mass Arbitration claims may proceed in any arbitration" (Character.AI).
• Replika: at 25 demands, each side picks five test cases, then a mediation of at least 60 days, then stages of up to 50 (Replika).

Arbitration also happens out of public view. Character.AI's and Replika's terms make "all aspects of the arbitration proceeding," and any award, "strictly confidential," and xAI's fallback arbitration is confidential too (xAI). A privacy claim settled that way leaves no public ruling for the next user to cite.

Microsoft's terms deserve credit on cost. For a claim under $75,000 outside a mass filing, Microsoft pays the filing and arbitrator fees, and if you win more than its last offer, it pays at least $1,000 plus reasonable attorney's fees (Microsoft).

Do these clauses apply in the EU and UK?

Mostly not. OpenAI's European terms, for the EEA, Switzerland, and the UK, say: "If we cannot resolve our dispute, you or we can go to your local courts" (OpenAI). Google's Irish version applies the law of your country and lets you file in your local courts (Google). Mistral moves EEA users to separate terms with home-country courts and voluntary mediation (Mistral), and Replika says its arbitration applies in the EU and UK "only to the extent permitted by your law" (Replika).

xAI sends European users to Irish courts under Irish law (xAI). And OpenAI's European terms still point EEA users to the EU's Online Dispute Resolution platform, which the European Commission says was "discontinued as of 20 July 2025" (European Commission).

What do these clauses mean for an AI privacy lawsuit?

They can decide whether one exists. On September 16, 2026, two ChatGPT users filed Vredenburgh v. OpenAI OpCo, a proposed class action in federal court in San Francisco. They allege that contractors at outside firms read users' conversations to grade ChatGPT's answers, and that OpenAI "never informed ChatGPT users" of it. They seek to represent "All persons in the United States who used ChatGPT." These are allegations, and no court has found them true. Our page on ChatGPT's Project Lily covers what the complaint says.

The 47-page complaint never mentions OpenAI's arbitration clause. Under OpenAI's terms, users who didn't opt out agreed that class actions "are prohibited." The complaint also asks for public injunctive relief, an order barring the practice for everyone, which OpenAI's terms leave to the San Francisco courts after arbitration. As of November 10, 2026, the docket shows no answer and no motion to compel arbitration, and the case was reassigned on November 5.

Against most of these apps, then, a user with a privacy claim who didn't opt out would likely bring it alone, before an arbitrator rather than a judge or jury. If many users have the same complaint, the batch and test-case rules above set the pace. A government agency can still act: Character.AI's and Replika's terms say the agreement doesn't stop you from taking issues to federal, state, or local agencies. For how stored chats themselves end up in court, see can AI chats be subpoenaed.

What a dispute could reach on Morse

Morse's own terms, last updated October 30, 2026, work much like most of the table. Section 18 asks you to email us first and try for 60 days to settle a dispute. After that it sends disputes to binding individual arbitration with the AAA, bars class actions, allows small claims court, has a mass-arbitration rule for 25 or more similar claims, and, to the extent the law allows, bars any claim not brought within one year after it arises. You can opt out within 30 days of first agreeing by emailing hello@trymorse.ai with your name and your account email.

Morse differs in what a dispute could reach. Every claim and discovery request in this article goes after records a company keeps, and the Vredenburgh complaint alleges that contractors read conversations OpenAI held. Morse keeps your chat history in your browser, encrypted, and our servers don't keep messages or log message content. A demand for your chats from an arbitrator, a court, or an agency would find a sign-in record, your plan and balance, and a usage ledger with no words in it. Because we don't store chats, we usually have very little to give. Morse also doesn't train AI on your content, sell your data, or show ads.

The privacy mode you pick for each chat decides who else could ever be asked for a message:

• Private, so no model host has a copy to produce. New chats start here, on open-weight models like DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers. When a provider confirms zero retention on a reply, Morse checks that confirmation and rejects any reply that lacks it.
• E2EE, for a chat Morse's servers can't hand over in readable form. Before a message leaves your device, your browser verifies the enclave's hardware proof and encrypts the message for that enclave alone, so Morse's servers only ever relay ciphertext. This depends on the enclave hardware and attestation working as designed. E2EE runs open-weight models, without web search or image generation, and is part of Pro, at $19 a month.
• Anonymous, so OpenAI, Anthropic, or Google gets the message under Morse's account, not yours. For GPT, Claude, and Gemini, our proxy removes your account, email, and IP address before the message goes out. The provider reads the text without knowing it's you and may keep it for a limited time to check for abuse, under its own policy, so keep identifying details out of what you type.

Your history locks when you sign out and takes two keys to open, one of which stays on your device. The copy on your device is the only one, so clearing your browser deletes it for good and Morse can't bring it back. Save a copy of anything you may need as evidence.

What to do this week

• ChatGPT, signed up in the last 30 days: use OpenAI's opt-out form if you'd rather keep your disputes in court.
• Replika, existing user: the opt-out in the October 28 terms runs 30 days from when they were posted. Email my@replika.com before then if you want out.
• Lumo or Perplexity: decide whether to opt out of the class waiver too, since dropping only the arbitration leaves the waiver in place.
• Any app: save a copy of the terms the day you sign up. Our page on AI privacy policy changes shows which apps keep old versions, and who owns your AI chats covers the license each one takes.

Try a chat from the Morse home page without an account, or start free with no card.