Is ChatGPT Safe? What OpenAI's Own Documents Show

Personal ChatGPT accounts share chats for model training by default, and OpenAI's settings let you turn that off. Deleted chats are generally scheduled for permanent deletion within 30 days, with…

Nov 16, 2026
•
6 min read
Technobezz
Is ChatGPT Safe? What OpenAI's Own Documents Showarda savasciogullari / Shutterstock.com

Contents

Don't Miss the Good Stuff

Get tech news that matters delivered to your inbox.

OpenAI's own documents show that ChatGPT collects what you type and upload, and that personal accounts share conversations for model training by default unless you turn that off. OpenAI says the content is encrypted and that it does not sell user data, and it has also disclosed several security incidents since 2023.

What ChatGPT collects

OpenAI's US privacy policy, dated September 10, 2026, lists account information such as names, contact details, birth dates, payment information and transaction history. It also covers the content you submit, including prompts, uploaded files, images, audio and video. Automatically collected data includes IP addresses, browser and device details, usage activity and country.

For Free and Go users, the policy adds advertising history and interests. OpenAI's consumer data help page says the company does not sell user data or share ChatGPT conversations with advertisers.

The same help page says authorized staff and service providers bound by confidentiality may access it to investigate abuse or security issues, for support, for legal matters, or to improve models unless you have opted out. OpenAI says that access is restricted, monitored and logged.

Training and the controls you have

  • Default: for personal Free, Plus and Pro workspaces, data sharing for model improvement is enabled by default.
  • Opt-out: turning off "Improve the model for everyone" under Settings, then Data controls, stops it while keeping chat history. OpenAI says opting out prevents new conversations from being used.
  • Feedback exception: on consumer plans, sending a thumbs up or down can allow that whole conversation to be used for training even with the opt-out on.
  • Temporary Chat: these chats stay out of history, do not create or update memories and are excluded from model improvement while they remain temporary. OpenAI may keep a copy for up to 30 days for safety.
  • Business plans: OpenAI says inputs and outputs from ChatGPT Business, Enterprise and Edu are not used for training by default.

What deleting a chat does

On a personal account, saved and archived chats remain until you delete them. Deleting a chat removes it from your account right away and schedules permanent deletion within 30 days, according to OpenAI's retention page. The exceptions are content that was already de-identified and content kept longer for security or legal reasons.

Saved memories are stored separately from chat history, so deleting a chat does not automatically delete them. A file saved separately in the Library also stays when its chat is deleted.

In an October 22, 2025 update on a litigation order, OpenAI said its obligation to retain new consumer ChatGPT content indefinitely ended on September 26, 2025. It said limited data from April to September 2025 would stay preserved for legal obligations.

Security measures and disclosed incidents

OpenAI states that consumer content is encrypted at rest and in transit. It also says its API and ChatGPT business services went through an independent SOC 2 Type 2 examination.

OpenAI's own posts describe these incidents:

  • March 2023: a bug in a software library let some users see chat titles from another active user's history. OpenAI said payment-related information for 1.2% of Plus subscribers active in a specific nine-hour window may have been exposed, that full card numbers were not, and that the bug was patched.
  • November 2025: a breach at analytics provider Mixpanel affected some API users and, per a December clarification, a limited number of ChatGPT users. Names, emails and approximate locations were among the data potentially affected. OpenAI said chats, passwords and payment details were not exposed.
  • August 2026: OpenAI reported that models in internal cybersecurity evaluations bypassed isolation controls and compromised research infrastructure and Hugging Face systems. It said OpenAI customer data was not affected.

What regulators have decided

Italy's data protection authority, the Garante, imposed a provisional restriction on OpenAI's processing of Italian users' data on March 30, 2023. It reported on April 28, 2023 that OpenAI had reinstated access after making changes, while its investigation continued. In December 2024 the Garante announced a 15 million euro fine. Its decision page now says that decision was temporarily removed after a Court of Rome judgment, published March 18, 2026, upheld the appeal against it.

In Canada, federal and provincial privacy regulators reported in May 2026 on GPT-3.5 and GPT-4, without assessing later models. They identified overcollection, a lack of valid consent and transparency, and inaccurate personal information in ChatGPT's initial development and deployment. The federal complaint was found well-founded and conditionally resolved, while British Columbia and Alberta found theirs well-founded and unresolved.

What you can do

If you do not want your chats used for training, turn off the model improvement setting and avoid the thumbs buttons, since OpenAI says feedback, and sessions in some opt-in programs it runs, can still be used for training. If a conversation is sensitive, Temporary Chat keeps it out of history and training, though a copy may be held for up to 30 days. OpenAI's terms also require users to be at least 13, with a parent's or guardian's permission under 18.

To clear old material, see how to delete ChatGPT conversations in bulk or how to delete a ChatGPT account. For the paid tiers, see Is ChatGPT Plus Worth It?

Share