Is ChatGPT safe? What OpenAI collects, who can see your chats and how to protect them

The short answer

OpenAI's own pages describe both risks and controls. As of October 22, 2026, OpenAI's privacy policy says it may use what you type to train the models behind ChatGPT, and that you can choose not to allow this. OpenAI's help center says a limited number of authorized OpenAI staff and trusted service providers can access conversations, for example to investigate abuse or handle legal matters.

OpenAI's help center describes a training opt-out, Temporary Chat, memory controls, chat deletion and multi-factor authentication. Its terms say ChatGPT's output may not always be accurate. OpenAI has published notices about security incidents in 2023 and 2025, described below.

As of October 28, 2026, ten state attorneys general have sued OpenAI over ChatGPT. Those are allegations, not findings.

What OpenAI collects

OpenAI's privacy policy, updated September 10, 2026, lists these kinds of personal data in its U.S. privacy policy:

• Account information, which can include payment details and transaction history.
• What you put into ChatGPT: prompts, uploaded files and data from services you connect.
• Messages you send OpenAI, for example by email, and information from surveys, events and identity or age checks.
• Log, usage and device data, such as your IP address, browser type and settings, the features you use and device identifiers.
• Your general location, worked out from your IP address. The policy says precise location is collected only if you choose to provide it.
• Cookies and similar technologies.
• For Free and Go users, information about your ad history and interests, and information from advertisers, such as purchases you make from them.

Are your chats used to train OpenAI's models?

They can be. As of October 22, 2026, OpenAI's help center says it may use content from Free, Plus and Pro accounts to train its models, and that by default it does not use content from ChatGPT Business, Enterprise or Edu workspaces or from its API. OpenAI says it takes steps to reduce the amount of personal information in training datasets.

To opt out, OpenAI's Data Controls FAQ says to turn off the setting called Improve the model for everyone. You can also choose Do not train on my content in OpenAI's Privacy Portal.

• Web: select your profile icon at the bottom left, then Settings, then Data Controls, and turn the setting off.
• iPhone: tap the three dots at the top right, then Settings, then Data Controls.
• Android: tap the three lines at the top left, then Settings, then Data Controls.
• OpenAI says the setting applies to your account, not one device, and that once it is off, new conversations are not used for training. Your chats stay in your history.
• OpenAI's help center says giving a thumbs up or thumbs down can make the whole conversation eligible for training, even if you have opted out.
• For Health in ChatGPT, OpenAI's help center says connected medical records and Apple Health data, and conversations that use them, are not used to train its foundation models, whatever your training setting.

Temporary Chat

OpenAI's help center describes Temporary Chat as a chat that stays out of your history and is not used to train its models while it remains temporary. To start one, open a new chat, select Temporary, and choose Personalized or Unpersonalized before you send the first message.

• Temporary chats do not create or update memories. A Personalized temporary chat can still read your existing memories and custom instructions; an Unpersonalized one does not use them.
• OpenAI says it may keep a copy of a temporary chat for up to 30 days for safety purposes, and that these chats are reviewed only when needed to monitor for abuse.
• If you save a temporary chat, it becomes a regular chat and follows your normal training and memory settings.

Memory and how to manage it

OpenAI's help center says ChatGPT's memory can draw on saved memories, which are details you ask it to remember or that it saves as useful, and on your past chats. The controls are under Settings, then Personalization, then Memory.

• To see or remove single saved memories, use the Manage or Saved memories control in Memory settings.
• To clear what ChatGPT remembers and switch memory off, open the more options menu in Memory settings and choose Delete and turn off memory. OpenAI says this does not delete your past chats.
• Deleting a chat does not necessarily delete a saved memory created from it. OpenAI says you need to delete both.
• OpenAI says it may keep logs of deleted saved memories for up to 30 days for safety and debugging.
• For personal accounts, OpenAI says chats and remembered information may be used to improve its models when Improve the model for everyone is on.

Who at OpenAI can see your conversations

As of October 22, 2026, OpenAI's help center says a limited number of authorized OpenAI personnel and trusted service providers, bound by confidentiality and security obligations, can access user content. It lists four reasons: investigating abuse or a security incident, helping you with account or product issues, handling legal matters, and improving model performance unless you have opted out. It says access is limited to those who need it and is logged.

In an August 26, 2025 post, OpenAI said conversations in which users appear to be planning to harm others are routed to a small team trained on its usage policies, which can ban accounts. It said it may refer cases involving an imminent threat of serious physical harm to others to law enforcement. In the same post it said it was not then referring self-harm cases to law enforcement, citing the private nature of ChatGPT conversations.

OpenAI's privacy policy also says it may disclose data to comply with a legal obligation, and that administrators of a business or Enterprise account may access and control accounts in that workspace. Its help center says content is stored on OpenAI's and its service providers' systems in the U.S. and other countries.

How long OpenAI keeps your data

OpenAI's help center says regular and archived chats stay saved to your account until you delete them. When you delete a chat, OpenAI says it schedules the chat for permanent deletion within 30 days, unless it was already de-identified and disassociated from your account, or OpenAI must keep it longer for security or legal reasons.

OpenAI's privacy policy lists cases where it may keep data after you delete it: content or accounts banned for violating its usage policies, data it is legally required to keep (it gives a lawful subpoena as an example), payment and transaction records, and a record of your deletion request.

Court-ordered retention OpenAI has disclosed

On June 5, 2025, OpenAI said a court order in The New York Times' copyright suit required it to keep consumer ChatGPT and API content indefinitely, including chats users had deleted. OpenAI said the order covered ChatGPT Free, Plus, Pro and Team and API customers without a zero data retention agreement, but not ChatGPT Enterprise or Edu.

In an October 22, 2025 update, OpenAI said those obligations ended on September 26, 2025, and that deleted chats and Temporary Chats are again removed within 30 days. It said it is still keeping a limited set of user data from April through September 2025, which it said the Times continues to demand that it keep. OpenAI said that data is accessible only to a small, audited legal and security team, can be used only to meet legal obligations, and would not be turned over to the Times, the court or anyone else at that time.

On November 12, 2025, OpenAI said the Times was seeking 20 million consumer ChatGPT conversations randomly sampled from December 2022 to November 2024. OpenAI said it was removing personal identifying information from them, holding them separately under a legal hold, and challenging the demand. On January 5, 2026, U.S. District Judge Sidney H. Stein in New York, in the consolidated copyright cases that include the Times' suit, denied OpenAI's objections and affirmed a magistrate judge's orders requiring OpenAI to produce a 20 million de-identified log sample to the news and class plaintiffs. He held that the magistrate judge had adequately balanced users' privacy interests, citing the reduced sample size, de-identification and an existing protective order. We have not confirmed later steps in the case.

Does OpenAI sell or share your data?

OpenAI's privacy policy says it does not sell personal data. It says it may share limited data with marketing partners to promote OpenAI's products to you on other websites, which some state laws treat as targeted advertising. You can opt out with the marketing privacy control in your account settings; when logged out, under Settings, then Data Controls; or with a Global Privacy Control signal.

The policy says OpenAI may share data with vendors and service providers, affiliates, a buyer or successor in a merger or similar deal, and government or others where legally required. OpenAI's help center says it does not sell your data or share your ChatGPT conversations with advertisers.

• Ads: OpenAI's help center says ads may appear for Free and Go users, starting with a U.S. test on February 9, 2026. Plus, Pro, Business, Enterprise and Edu accounts do not have ads.
• OpenAI says advertisers do not get access to your chats, chat history, memories or personal details, and receive aggregated data such as views and clicks.
• OpenAI says ads are not shown near sensitive topics such as personal health, mental health or politics, or to accounts identified as under 18.
• Under Settings, then Ad Controls, you can turn off ad personalization or delete ads data. OpenAI says deleted ads data can take up to 30 days to be removed from its servers.

Security features and incidents OpenAI has disclosed

OpenAI's help center says ChatGPT chats are encrypted at rest and in transit. Its privacy policy says no internet transmission is fully secure and advises care about what you share.

• Multi-factor authentication: OpenAI's help center says to open Settings, then Security, and turn on Multi-factor authentication. It supports authenticator-app codes, with a recovery code or an emailed code as fallbacks. The same page describes a Log out of all devices option.
• Locked chats: OpenAI's Privacy Center article lists locked chats, protected by an account PIN, where available.
• March 2023: OpenAI said a bug in an open-source library let some users see titles from another active user's chat history. It said payment-related details of 1.2% of ChatGPT Plus subscribers active during a nine-hour window on March 20, 2023 may have been visible, including name, email, payment address, card type, the last four card digits and expiration date, but not full card numbers.
• November 2025: OpenAI said an attacker gained access to part of the systems of Mixpanel, an analytics vendor, and exported a dataset. OpenAI said this was not a breach of its own systems. It said the data could include names, email addresses, approximate location and browser details of API platform users and, per a December 19, 2025 clarification, a limited number of ChatGPT users who had submitted help tickets or were logged into platform.openai.com. OpenAI said no chats, passwords, API keys, payment details or government IDs were exposed, and that it stopped using Mixpanel. It recommended turning on multi-factor authentication and watching for phishing.

Accuracy and advice limits

OpenAI's Terms of Use, effective January 1, 2026, say output may not always be accurate, that you should not rely on it as a sole source of truth, and that you must evaluate it, using human review as appropriate. Its help center says ChatGPT can produce incorrect or misleading answers, can sound confident when wrong, and can invent citations, and tells users to check important information.

OpenAI's usage policies, effective October 29, 2025, bar using its services for tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional. They also bar automating high-stakes decisions in sensitive areas, including credit, employment, housing, insurance, legal and medical matters, without human review. OpenAI's help center says Health in ChatGPT is not intended for diagnosis or treatment.

Teens and children

OpenAI's terms say users must be at least 13, or the minimum age required in their country, and that users under 18 need a parent's or guardian's permission. OpenAI's help center says ChatGPT estimates whether an account may belong to someone under 18 and, if so, applies teen protections and shows no ads. OpenAI's parental controls article says linked parents can change some settings, including whether a teen's chats are used for training, but cannot read or monitor the teen's conversations. Our guide to whether ChatGPT is safe for kids covers age checks, teen limits and parental controls in detail.

The state attorney general lawsuits

Florida's attorney general sued OpenAI and its CEO, Sam Altman, on June 1, 2026. On October 14, 2026, Montana's attorney general announced a suit and said Arizona, Arkansas, Idaho, Iowa, Kansas, Nebraska and West Virginia had also sued OpenAI in their own courts. Among other claims, Florida alleges ChatGPT collects data from minors without meaningful parental oversight, and Montana alleges ChatGPT does not verify users' ages. These are allegations, not findings. Our guide to ChatGPT lawsuits lists the cases and their status.

Which laws may apply

No law names ChatGPT. Whether ChatGPT counts as a companion chatbot or AI companion under a particular state law is a legal question. These laws, bills and government actions deal with some of the issues in this guide:

• California SB 243, in effect since January 1, 2026: according to its author's office, companion chatbot operators must keep a protocol for responding to suicidal ideation, suicide and self-harm that refers users to crisis services, and must remind users it knows are minors that they are talking to AI; a later law, SB 1119, removes those known-minor duties on January 1, 2027. A person injured by a violation can sue.
• New York's AI companion law (General Business Law Article 47): operators must have a protocol to detect and respond to expressions of suicidal thoughts or self-harm and refer users to crisis services, and must tell users at the start of an interaction, and at least every three hours, that they are not talking to a human.
• The FTC ordered seven companies, including OpenAI OpCo, to report on their companion chatbots, including how they use or share personal information from users' conversations (September 11, 2025). It is a study, not an enforcement case.
• 42 attorneys general, led by Pennsylvania, wrote to 13 AI companies, including OpenAI, in December 2025, asking for safety testing, recall procedures and clear warnings to consumers.
• The People-First Chatbot Act (H.R. 9619), introduced July 9, 2026, would bar chatbot providers from selling chat logs, require users' affirmative consent before training on adults' chats, bar training on minors' chats and require deletion on request. It is pending and is not law.
• The ten state attorney general suits against OpenAI each have their own record on this site, linked below.

We are not saying OpenAI is covered by, or out of compliance with, any of these laws. It may have practices it has not published, and whether a law applies to it is a legal question this guide does not decide.