Is ChatGPT Safe for Your Privacy? What AI Chatbots Do With the Data You Share

Is ChatGPT safe to use for private information?

ChatGPT and similar chatbots are generally safe for everyday tasks, but they are not private in the way a conversation with a doctor or lawyer is. Your messages are kept by the provider, may be used to improve its models unless you opt out ^[4], and can be subject to legal demands ^[5]. A Stanford Institute for Human-Centered AI (HAI) review of 28 policy documents from Amazon, Anthropic, Google, Meta, Microsoft and OpenAI found that all six use chat data to train their models by default, and that some keep chat data indefinitely ^[1].

OpenAI's chief executive Sam Altman has acknowledged the gap. In July 2025 he said that conversations with a therapist, lawyer or doctor carry legal privilege, adding: 'we haven't figured that out yet for when you talk to ChatGPT' ^[10]. According to TechCrunch, Altman added that this could create a privacy concern in a lawsuit, because OpenAI would be legally required to produce those conversations ^[10].

The short answer

• Turn off model training if you do not want your chats used for it ^[2][7][8].
• Use temporary or activity-off modes for one-off sensitive questions, knowing they are still kept for a short period ^[3][8].
• Do not enter passwords, ID numbers, medical records or confidential work data; Google itself asks users not to enter confidential information they would not want a reviewer to see ^[8].

What happens to the data you share with an AI chatbot?

Several different things can happen to a conversation, and they are controlled separately.

Model training

Training means your conversations may be used to improve future versions of the model. OpenAI says ChatGPT improves by further training on the conversations people have with it, unless users opt out ^[4]. Stanford's researchers warn that sensitive information shared in chats, including in uploaded files, may be collected and used for training ^[1].

Human review

Some companies let people read conversations. Google states that a subset of Gemini chats is reviewed by human reviewers, including trained service providers, to improve its services ^[8]. The Stanford study found that some developers allow humans to review chat transcripts for training ^[1].

Retention

Retention is how long the company keeps your data, which can differ from what you see in your chat history. For example, Google keeps Gemini chats made with Keep Activity turned off with your account for 72 hours ^[8], and OpenAI may keep a copy of a ChatGPT temporary chat for up to 30 days for safety purposes ^[3].

Sharing and third parties

Shared chat links are a separate exposure, which ChatGPT lets you manage in Data controls ^[2]. When a custom GPT sends data to an outside service through an 'action', that data is governed by the recipient's privacy policy and may be kept for longer than 30 days ^[3].

Does ChatGPT use your conversations for training?

Yes, by default, for personal accounts. OpenAI's help centre says that on personal workspaces with ChatGPT Free, Plus or Pro, data sharing is enabled by default ^[2]. To stop new conversations being used for training ^[2][4]:

1. On the web, open your account menu and select Settings, then Data controls ^[2].
2. Turn off Improve the model for everyone and select Done ^[2].
3. On iOS or Android, open the sidebar, select your profile icon, choose Data controls and turn off the same setting ^[2].
4. Alternatively, use OpenAI's privacy portal and choose 'do not train on my content' ^[4].

Some details are easy to miss. Voice mode has a separate control for sharing data for training, and turning off training does not delete or hide saved chats ^[2]. Archiving a chat does not change whether it can be used for training ^[2]. If you give feedback, such as a thumbs up or thumbs down on a response, OpenAI says the entire conversation may be used for training even after you opt out ^[2].

What about temporary chat?

Temporary chats do not appear in your history, do not create or update memories and are not used to improve OpenAI's models ^[3]. OpenAI may still keep a copy for up to 30 days for safety purposes ^[3]. If you save a temporary chat, it becomes a regular chat and follows your normal settings ^[3].

Work accounts

OpenAI says that by default it does not use content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces, or from its API, to train its models ^[2]. Your employer's own retention and access policies still apply ^[2].

How do Gemini and Claude handle your data?

Google Gemini

When Keep Activity is on, Google uses your activity to provide, develop and improve its services, including training generative AI models; activity is auto-deleted after 18 months by default, and you can change this to 3 or 36 months or choose not to auto-delete ^[8]. With Keep Activity off, future chats are not used to train AI models unless you send feedback, but they are retained with your account for 72 hours ^[8]. Chats that human reviewers have looked at are disconnected from your account and kept for up to three years ^[8].

Anthropic Claude

On 28 August 2025, Anthropic changed its consumer terms for Claude Free, Pro and Max, including Claude Code used with those accounts ^[7]. Users were asked to choose whether their chats could be used for model training, with a deadline of 8 October 2025 for existing users ^[7]. If you allow training, data retention extends to five years; if you do not, the existing 30-day retention period applies ^[7]. Deleted conversations are not used for future training, and the choice can be changed in Privacy Settings at any time ^[7]. Commercial services such as Claude for Work and API use are excluded ^[7]. Stanford's researchers describe the change as training on Claude conversations by default unless users opt out ^[1].

The table below summarises the consumer defaults as described by each company, as of 3 November 2026. Settings and names change often, so check the current help pages before relying on them.

Can deleted chats still be kept or handed over?

Deleting a chat normally removes it after a set period, but legal obligations can override that. In May 2025, a court in the New York Times' copyright lawsuit against OpenAI ordered the company to preserve ChatGPT logs, including chats users had deleted ^[5][6]. OpenAI says its obligations under that order ended on 26 September 2025, and that it has returned to its standard practice of deleting removed chats and temporary chats within 30 days ^[5]. Engadget reported that a new order filed by federal judge Ona T. Wang on 9 October 2025 ended the obligation to preserve such data going forward, but that OpenAI must still keep data for ChatGPT accounts flagged by the Times ^[6].

OpenAI also says it continues to store limited historical user data from April to September 2025 under legal hold, accessible only to a small audited legal and security team, and that it is no longer required to retain conversations originating from the European Economic Area, Switzerland or the United Kingdom ^[5]. The case illustrates a broader point: chat histories are business records that can become evidence ^[10].

Accidental public sharing

Sharing settings can also expose chats. In 2025 OpenAI briefly offered an opt-in checkbox that let shared ChatGPT conversations be indexed by search engines such as Google ^[9]. OpenAI's chief information security officer, Dane Stuckey, called it a short-lived experiment and said it was removed because it 'introduced too many opportunities for folks to accidentally share things they didn't intend to' ^[9]. OpenAI said some indexed conversations might stay visible for a while because of search-engine caching ^[9].

What have privacy regulators done about AI chatbots?

Regulators have tested how data protection law applies to chatbots, with mixed outcomes. On 20 December 2024, Italy's data protection authority, the Garante, announced a €15 million fine against OpenAI ^[11]. It found that OpenAI had not notified the authority of a data breach in March 2023, had used personal data to train ChatGPT without first identifying an adequate legal basis, had breached transparency obligations and lacked age verification ^[11]. It also ordered a six-month public information campaign ^[11].

The decision did not stand unchallenged. The Garante's own page notes that the Court of Rome, in judgment No. 4153/2026 published on 18 March 2026, upheld OpenAI's opposition to the decision, which was temporarily removed from the authority's website ^[11]. As of 3 November 2026, the fine should therefore be treated as annulled by the Rome court, subject to any further appeal.

In the United States, the Stanford researchers describe a patchwork of state-level laws and a lack of federal regulation, and they call for comprehensive federal privacy regulation, opt-in consent for model training and default filtering of personal information from chats ^[1].

What should you never share with an AI chatbot?

A useful rule is to share nothing you would not want a stranger at the company, or a court, to read ^[8][10]. In practice, that means avoiding:

• Passwords, one-time codes and recovery phrases. No chatbot needs them to help you.
• Government ID and bank details, including passport, national ID and card numbers.
• Medical and mental-health details tied to your name. Stanford notes that health and biometric data shared in chats may be collected and used for training ^[1], and Altman has said such conversations lack legal privilege ^[10].
• Confidential work material, unless your employer provides a business account that excludes training by default ^[2].
• Other people's personal information, such as a friend's diagnosis or a colleague's contact details.

Safer habits

• Remove names and identifying details before pasting documents.
• Use temporary chat or turn off activity for one-off sensitive questions ^[3][8].
• Review and delete old conversations, and check which chats you have shared as links ^[2][9].
• Check children's use: Stanford found most developers do not remove children's input from training ^[1].

A note for readers in Iran

As of 3 November 2026, Iran does not appear on OpenAI's list of countries and territories where it supports access to ChatGPT, and OpenAI warns that accessing its services from outside the listed places may result in an account being blocked or suspended ^[12]. Accounts reached through workarounds may therefore be lost along with their chat history, so keep your own copies of anything important.

What are the trade-offs of chatbot privacy settings?

What the settings protect

• Opting out stops new ChatGPT conversations being used for training ^[4].
• Turning off Gemini's Keep Activity stops future chats being used to train AI models unless you send feedback ^[8].
• Declining training on Claude keeps the 30-day retention period instead of five years ^[7].

What they do not protect

• Short-term retention for safety still applies, for example up to 30 days for ChatGPT temporary chats ^[3] and 72 hours for Gemini ^[8].
• Opting out does not remove data from models already trained; Anthropic says data used in training that has started, or in models already trained, stays there ^[7].
• Court orders can override deletion, as the 2025 preservation order showed ^[5].
• Chats reviewed by people at Google are kept for up to three years ^[8].

What happens next for AI chatbot privacy?

Policies are changing fast. Since mid-2025, Anthropic moved its consumer users to an explicit training choice ^[7], OpenAI ended a court-ordered retention period ^[5] and removed a search-indexing option ^[9], and an Italian fine was overturned in court ^[11]. As chatbots add features such as memory and file libraries ^[2][3], the amount of personal data they hold grows, which makes defaults more important.

What remains uncertain

• Legal privilege for AI conversations. Altman has said the question is unresolved ^[10].
• The outcome of the New York Times lawsuit and what happens to the preserved April to September 2025 data ^[5].
• Whether regulators will require opt-in consent for training, as the Stanford researchers recommend ^[1].

This is general information, not legal advice. For decisions about sensitive data, check each provider's current privacy policy and, where relevant, your national data protection authority.