Claude is among the more privacy-respecting mainstream assistants. Anthropic does not train on consumer conversations by default, retains chats while your account is active, and deletes conversations within 30 days of you deleting them. Flagged conversations can be kept up to two years for trust and safety review.
Short answer
Claude is among the more privacy-respecting mainstream assistants. Anthropic does not train on consumer conversations by default, retains chats while your account is active, and deletes conversations within 30 days of you deleting them. Flagged conversations can be kept up to two years for trust and safety review.
Step by step
Check your training setting: Open Settings, then Privacy, and confirm whether the option to help improve Claude is on. Anthropic has changed this default over time, so verify rather than assume.
Review what you have shared: Scan your conversation list for anything containing client data, credentials, or personal identifiers, and delete those threads.
Export before cleaning up: Settings, Privacy, Export Data produces a JSON archive by email. Get it before you delete conversations, because deletion is not reversible.
The short version
Anthropic's consumer privacy posture is stronger than most competitors. The company's public position has been that conversations from consumer products are not used to train models unless you opt in or your conversation is flagged for safety review.
That is a meaningfully better default than "on unless you find the toggle."
What is still true
Anthropic stores your conversations. They sit on Anthropic infrastructure for as long as your account exists.
Deletion is a 30 day process. Deleted conversations go from your view immediately and are removed from backend systems within 30 days.
Flagged content is different. Conversations that trip trust and safety systems can be retained up to two years.
Employees can access data under limited circumstances, for abuse investigation and legal compliance.
Where the real risk is
For most people the risk with Claude is not Anthropic, it is themselves. Pasting a client contract, a patient summary, or a production credential into any hosted assistant creates a copy you do not control, and no privacy policy makes that copy disappear on demand.
Redaction beats policy every time.
The other half of safety
Safety also means not losing work. Claude has no offline mode and no local archive. If your account lapses, gets locked, or you delete a thread by accident, the conversation is gone.
Export a JSON archive periodically and convert it into readable documents. That way, your reasoning, drafts, and research live in files you own rather than in an account.
Frequently asked questions
Does Anthropic train on my Claude conversations?
Not by default for consumer accounts, unless you opt in or the conversation is flagged for safety review. Verify the current setting under Settings, Privacy.
How long does Claude keep deleted chats?
Deleted conversations are removed from backend systems within 30 days. Flagged conversations can be kept up to two years.
Is Claude safe for confidential work?
Commercial tiers offer stronger contractual protections. On a consumer account, redact identifiers before pasting anything confidential.
Can I get a copy of everything Claude has stored?
Yes. Settings, Privacy, Export Data emails you a JSON archive of your conversations.