Is ChatGPT safe? What it keeps, and what never to tell it

Yes, with limits. ChatGPT is a real service from OpenAI, and its main risks come from what you type, not from a virus. Your chats are stored, consumer chats may be used to train OpenAI’s models unless you opt out, and a court can order OpenAI to keep them. So never type passwords, Social Security numbers, health records, or work secrets.

What ChatGPT keeps

The App Store page for the ChatGPT iPhone app (read October 27, 2026) shows what OpenAI says it handles. Apple notes that it has not verified this label. It lists these kinds of data “linked to you”:

• Contact info: name, email address and phone number.
• User content, including audio data, and search history.
• Identifiers, usage data and diagnostics.
• Coarse location, and health and fitness data.

It names purposes such as app functionality, analytics, product personalization, and advertising or marketing. The developer is listed as OpenAI OpCo, LLC.

Training. According to TechCrunch (September 25, 2026), these are OpenAI’s rules. Enterprise users are automatically opted out of having their chats used to train future models. Consumer users are opted in unless they choose not to share their data. Even then, clicking thumbs-up or thumbs-down on a conversation still makes that interaction available for training.

History and memory. Regular chats show up in your chat history. TechCrunch (October 20, 2026) says Temporary Chats do not. ChatGPT also has a Memory feature. Texas’s petition quotes OpenAI’s Help Center as describing it as “remember[ing] relevant preferences and details from your chats and other available sources.”

Deleted chats and courts. A deleted chat may not be gone if a court requires OpenAI to keep it. On May 13, 2025, a federal magistrate judge in New York ordered OpenAI to preserve output log data “that would otherwise be deleted.” That covered data deleted at a user’s request or under privacy laws. The order came in the copyright case brought by The New York Times and others. On October 9, 2025, the court approved an agreement ending the duty to keep new deleted data as of September 26, 2025, with exceptions. OpenAI must still keep the data it had already saved under the order, except data from users in the European Economic Area, Switzerland and the UK.

In January 2026, a federal judge upheld an order for OpenAI to hand over a sample of 20 million chat logs. According to the ABA Journal (January 8, 2026), OpenAI would remove identifying details first, and a protective order applies. OpenAI argued the order would invade users’ privacy, and the judge rejected that argument.

Settings you can use

These come from TechCrunch’s reporting on ChatGPT in 2025 and 2026. Look in ChatGPT’s settings for them, since menu names change.

• “Improve the model for everyone.” This setting controls whether your chats help train OpenAI’s models. Remember the thumbs-up and thumbs-down rule above.
• Temporary Chat. It keeps a chat out of your history. Do not assume it is private.
• Locked Chats. OpenAI began rolling this out on October 20, 2026. You lock chats with a four-digit PIN, Face ID or Touch ID. OpenAI says locking is not end-to-end encryption. It does not change how long a chat is kept, or the “Improve the model for everyone” setting. It helps when someone else uses your screen or device.
• Memory. It is a setting, and TechCrunch (September 29, 2025) says parents can turn it off for a teen.

The news: lawsuits and teen safety

An allegation in a lawsuit is a claim, not a finding. Here is what officials have said and filed.

• October 14, 2026: Nebraska’s attorney general announced a lawsuit against OpenAI. His release says the attorneys general of Iowa, Kansas and West Virginia joined him, that a bipartisan group of states sued OpenAI individually, and that Arizona, Arkansas, Idaho and Montana have also sued in their own state courts. Arizona’s attorney general alleges OpenAI engineered ChatGPT to maximize compulsive use among minors, and that risks were not disclosed to Arizonans. OpenAI told AZFamily: “While no system can be perfect, we’ll continue to work with experts to inform our work.” It pointed to parental controls and break reminders.
• October 23, 2026: Texas’s attorney general sued OpenAI and its CEO, Sam Altman, in Collin County. The release says OpenAI marketed ChatGPT as a safe and reliable tool but failed to protect Texans, including children. The claims are under the state’s Deceptive Trade Practices law. The petition also cites the SCOPE Act, a Texas law on protecting minors online.
• September 11, 2025: The FTC ordered seven companies, including OpenAI, to report on how their companion chatbots affect children and teens, and how they handle user data. The FTC said such studies have no specific law enforcement purpose.

On teen safety, TechCrunch reported that OpenAI added parental controls on September 29, 2025. Parents can set quiet hours, turn off voice mode and memory, remove image generation, and opt out of model training. Flagged chats go to trained reviewers, and OpenAI may alert parents if it sees signs of acute distress. In August 2026, OpenAI announced ChatGPT for Teens. TechCrunch says age-appropriate protections are on by default there.

Has ChatGPT been hacked or leaked?

Here is what has been reported, with dates.

• Shared links, July 2025. Shared ChatGPT chats showed up in Google results. OpenAI said chats “are not public unless you choose to share them.” It removed the option that let users make shared chats discoverable by search engines. Treat any share link like a public post.
• Mixpanel, November 2025. A breach at an analytics vendor exposed names, email addresses, approximate city and state, and some device data. OpenAI said it did not directly affect ChatGPT users. TechCrunch said the people affected were likely developers who use OpenAI’s products (December 2, 2025).
• Code-library attack, May 2026. OpenAI said two employees’ devices were affected by an attack on a popular open-source library. It said it found no evidence that user data was accessed (TechCrunch, May 14, 2026).
• Research agents, September 2026. TechCrunch reported that OpenAI said AI agents in its research environment posted 53 user-provided images, which had been in training data, to image-hosting sites. OpenAI said the links were not publicly listed. OpenAI called this “not an appropriate use of this data.”

What never to type

The UK’s National Cyber Security Centre (NCSC), in a post dated March 14, 2023, says a query is visible to the company that runs the model. It may also be used to develop later versions. The NCSC advises people not to include sensitive information in queries to public chatbots. It also says stored queries may be hacked, leaked, or accidentally made public. Never type:

1. Passwords, one-time codes, or security answers.
2. Social Security, driver’s license, passport, or bank and card numbers.
3. Health records or a diagnosis tied to your name.
4. Private details about other people.
5. Client, employer, or school data you are not allowed to share.
6. Photos or documents that show personal details.

Work plans differ. TechCrunch reports that OpenAI opts enterprise users out of training automatically. Still, follow your employer’s rules before you paste anything.

Fake ChatGPT apps and extensions

Scammers use popular names as bait. In May 2023, Meta said it had found ten malware families using ChatGPT-themed lures. It described malicious browser extensions in official web stores that claimed to offer ChatGPT-based tools (TechCrunch, May 3, 2023). To avoid fakes, open the service by typing the address yourself or using the official app. On an iPhone, the App Store listing names the developer as OpenAI OpCo, LLC. Be wary of any extension that asks for your logins.

What can go wrong

• Deleting a chat may not erase it if a court order or law requires it to be kept.
• Turning off training does not stop the thumbs-up and thumbs-down rule.
• A chat shared by link can be seen by anyone who gets the link.
• Parental controls do not replace a conversation. Talk with your teen directly.
• These rules and settings change. Check ChatGPT’s settings again after major updates.

What to do now

1. Decide what you will never type, using the list above.
2. Open ChatGPT’s settings and decide on “Improve the model for everyone.” Skip thumbs ratings on private chats.
3. Use the official app or website only, and check the developer name before you install.
4. If your teen uses ChatGPT, set up parental controls together and talk about what is safe to share.
5. For scams that use AI, see AI scams. To protect your accounts, see Am I hacked?