AI chat privacy

Deleted is not always deleted.

Millions of people now ask AI chat apps about their divorce, their immigration case, their arrest or their lawsuit. Here is what the three largest consumer apps say, in their own published policies, about what they keep, what they flag, who can read it and when a court can get it.

On consumer AI apps, your legal questions are not privileged. Automated systems scan them. A flagged chat can be read by staff, kept for years and reported to the police. A lawsuit or court order can require any chat to be preserved and handed over, deleted or not. Your settings do not change this.

What each company says

Claude (Anthropic), consumer plans

Deleted chats: removed from back-end storage within 30 days.
Flagged chats: when automated trust and safety systems flag a chat for a usage policy violation, its inputs and outputs are kept for up to 2 years, and its trust and safety classification scores for up to 7 years. Deleting the chat does not shorten this.
Human review: content flagged by the automated systems can be reached by a small designated review team; Anthropic says every read is logged.
Legal requirements: data can be kept where Anthropic is legally required to keep it. No time limit is stated.
Scale: in the first half of 2026 Anthropic reported 11.4 million banned accounts and 15,079 reports to the National Center for Missing and Exploited Children.

Sources: How long do you store my data?, Data retention practices for covered models, Using Claude for legal work, Transparency Hub.

ChatGPT (OpenAI)

Deleted chats: scheduled for permanent deletion within 30 days, unless OpenAI must keep them longer for security or legal obligations. No time limit is stated for those exceptions.
Temporary chats: a copy may be kept for up to 30 days for safety purposes.
Flagged chats: no fixed retention period is published for content flagged for policy violations, fraud, safety or security.
Police: after the February 2026 school shooting in British Columbia, OpenAI told the Canadian government it would adopt more flexible standards for reporting accounts to law enforcement.
Court orders: in the New York Times copyright case, a court ordered OpenAI on 13 May 2025 to preserve consumer ChatGPT logs that would otherwise have been deleted, including chats users had deleted, across Free, Plus, Pro and Team. That obligation ended on 26 September 2025. On 7 November 2025 the court ordered 20 million de-identified ChatGPT logs produced to the plaintiffs.

Sources: Chat and file retention policies in ChatGPT, Gigazine, terms.law, iTechGuides.

Gemini (Google)

Human reviewers: a subset of conversations is read by human reviewers, including trained service providers.
Reviewed chats: kept for up to three years, and not deleted when you delete your activity.
With activity off: conversations are still kept with your account for up to 72 hours.
Google's own warning: do not enter confidential information that you would not want a reviewer to see.

Source: Gemini Apps Privacy Hub.

Why this matters for legal questions

Flags are triggered by what you write, not by who you are. The topics that trip automated safety systems are violence, weapons, drugs, fraud and self harm. Someone explaining their own criminal charge, an abusive marriage or how a scam was run on them has to describe exactly those things. No company publishes what triggers a flag, so there is no way to know whether a chat has been flagged.

Even an unflagged chat is a risk while it exists. In 2026 a federal court held that a person's conversations with a consumer AI tool were not protected by attorney-client privilege (United States v. Heppner). A chat that is not privileged can be demanded in a lawsuit or an investigation, and a court can order it kept no matter what the deletion schedule says.

Common questions

Can AI companies keep my chats forever?

For chats flagged by automated safety systems, Anthropic keeps inputs and outputs for up to 2 years and safety classification scores for up to 7 years. OpenAI publishes no fixed limit for flagged content. Every major provider keeps an open-ended right to retain data for legal obligations, which lasts as long as the legal matter does.

Can they keep a chat for any reason?

Not any reason, but the stated reasons are broad: safety, security, fraud and legal obligations. A subpoena, a lawsuit or a government request is enough.

Does deleting an AI chat protect me?

It shortens the normal window to about 30 days. It does not undo a flag, a human review, a legal hold or a court order.

Do business AI plans work differently?

Often yes. Enterprise and API customers can negotiate shorter retention, including zero data retention. Consumer apps follow the published consumer policies.

Are my AI chats about a legal problem privileged?

Not on a consumer AI app. In United States v. Heppner (2026) a federal court held that a person's conversations with a consumer AI tool were not protected by attorney-client privilege.

How Confidelio is different

Research runs on AI hosted inside Confidelio's own Amazon Web Services account with data retention set to none for the Claude models that do the research, and the model maker never receives what you write.
Your name and the names of the people in your case are replaced with neutral labels before anything reaches the AI.
Your cases are encrypted with a key only your password unlocks.
Confidelio is built to retain attorney-client privilege from the first chat.

One limit applies to everyone, us included: Amazon screens for child sexual abuse material and may keep and report it. Read the full privacy page.

Create your account

Related guides

Every statement about another company above comes from that company's published policy or the reporting linked beside it, as checked on 22 October 2026. Policies change; follow the links for the current versions. This page is information, not legal advice.