ChatGPT

Is ChatGPT safe?

ChatGPT will not steal your bank details, and for everyday questions it is a reasonable risk. The catch is that it remembers. OpenAI keeps your conversations, uses them to train its models unless you switch that off, can have staff review them, and since February 2026 shows ads to free users in the US. Regulators in Italy and Canada have found problems with how it handled personal data, and families have sued over its answers to teenagers in crisis. We score it 57/100.

The way to use it safely is to know what you are sharing and change two settings.

What OpenAI keeps

Every chat is stored on OpenAI's servers, encrypted at rest but not end to end, so the company can access it. On Free, Go and Plus plans, conversations are used to improve models by default. Apple's App Store label lists health and fitness data, coarse location, name, email, phone number, audio, other user content, search history, identifiers, usage and diagnostics as linked to you.

People tell ChatGPT things they would not tell a search engine: symptoms, money worries, relationship problems. That is the reason to be careful. In 2025 a court order in the New York Times copyright case forced OpenAI to preserve users' deleted chats for several months, a reminder that "deleted" depends on the law as well as on OpenAI. And in mid-2025 thousands of chats people had shared with a public link turned up in Google results after users ticked a "make discoverable" box. OpenAI removed the option.

Ads arrive

On 9 February 2026 OpenAI began showing ads to logged-in adults on the Free and Go plans in the US. Ads appear below an answer when a sponsored product matches the topic of the conversation and are labelled as sponsored. OpenAI says ads do not influence answers and are not shown to minors. Plus and Pro subscribers see none. It is a significant change for an app that holds so many personal conversations, and the reason privacy scores 45 rather than higher.

Security record

The incidents so far have been bugs and vendor problems, not a breach of OpenAI's core systems. In March 2023 a software bug let some users see other users' chat titles, and exposed partial payment details of about 1.2% of Plus subscribers active during a nine-hour window. In November 2025 OpenAI said analytics vendor Mixpanel had been breached, exposing names and email addresses of some API platform users, not ChatGPT conversations. Multi-factor authenticationTwo-factor authenticationSigning in with a second proof of identity on top of your password, such as a code from an app, a security key or a passkey.More in the glossary is available and worth turning on.

Teenagers and harmful answers

In August 2025 the parents of Adam Raine, a 16-year-old who died by suicide in April 2025, sued OpenAI and Sam Altman, alleging ChatGPT discussed suicide methods with him and presented itself as his closest confidant. More families filed similar lawsuits later that year. OpenAI denies responsibility and says its safeguards can weaken in very long conversations. None of these cases had gone to trial by October 2026.

OpenAI responded with parental controls in September 2025, letting parents link a teen's account, set quiet hours, turn off memory and get notified of signs of acute distress. These help, but only if the teen uses the linked account.

For adults the bigger everyday risk is error. ChatGPT can give confident, wrong answers on medication doses, legal deadlines or tax rules. Check anything important with a professional or an official source.

Regulators

Italy's data protection authority fined OpenAI €15 million in December 2024 for training on personal data without an adequate legal basis, poor transparency and no age verification. In March 2026 the Court of Rome annulled the fine on procedural grounds, because Ireland had become OpenAI's lead EU regulator, not because it found the practices lawful. On 6 May 2026 Canada's federal and provincial privacy commissioners published joint findings that OpenAI had over-collected personal information, lacked valid consent and fell short on accuracy, access and deletion.

On 30 September 2026 an FTC spokesperson confirmed that the agency is investigating OpenAI, Anthropic and other AI companies over risks their technology may pose to consumers, without giving details. The probe follows OpenAI's disclosures that its models had acted beyond their instructions, including interacting with US government websites. An investigation is not a finding of wrongdoing; the FTC has not said whether it suspects any breach of the law.

Settings worth changing

1. Settings > Data controls: turn off "Improve the model for everyone".
2. Use Temporary Chat for anything sensitive. It is not saved to history or used for training.
3. Review and delete what ChatGPT remembers about you under Personalization > Memory.
4. Turn on multi-factor authenticationTwo-factor authenticationSigning in with a second proof of identity on top of your password, such as a code from an app, a security key or a passkey.More in the glossary in the security settings.
5. Do not paste passwords, ID numbers or confidential work files.

How we scored it

Transparency and control (72) is ChatGPT's strongest factor thanks to clear opt-outs, and security (64) is reasonable. Privacy (45) is held down by default training and the new ads, and scams and harmful content (50) by the teen-safety lawsuits. Video calls raise a different set of risks; see Zoom.