Skip to content

Anthropic (Claude)

The maker of Claude — and the host for anything you share as a Claude Artifact. For most things you build and share here, the honest answer is fine: nothing you publish as an Artifact is secret, and the data questions below are about your private chats, not the thing you hand out.

The one knob worth knowing: on a personal plan, your chats and coding sessions train future Claude models unless you switch that off — a single toggle, named below.

Last verified: 2026-06-07 · Confidence: high on the training toggle, retention, and the free-vs-enterprise split (all from Anthropic's own docs); data residency is the soft spot.

Does it train on what you type?

Personal plans (Free, Pro, Max — including Claude Code on those accounts): yes, by default, unless you turn it off. Since the August 2025 terms update, new chats and coding sessions feed model training when the Help Improve Claude setting is on. [confirmed] (consumer terms update)

  • Turn it off in one toggle. Settings → Privacy → Help Improve Claude → off (or go straight to claude.ai/settings/data-privacy-controls). Takes effect immediately, applies to new and resumed chats. [confirmed] (how to change it)
  • Off ≠ erased from past models. Anything already used in a training run that's started stays in — switching off only stops future runs. [confirmed]

Commercial plans (Team, Enterprise, Claude for Work / Government / Education, and the API): no training by default — there's no toggle to police, it just doesn't happen unless you deliberately opt in (e.g. the Development Partner Program). [confirmed] (is my data used for training?)

Retention and deletion

  • Delete a conversation and it's gone from training. A deleted chat is never used for future training, and is purged from the back end within ~30 days. [confirmed] (consumer terms update)
  • How long the rest is kept: with training on, up to 5 years; with it off, the standard 30-day window. [confirmed]
  • Two carve-outs that survive deletion: content your feedback (thumbs up/down) is attached to is kept up to 5 years (de-identified first); a conversation flagged by safety classifiers is held longer for policy enforcement. [confirmed] (is my data used for training?)

What enterprise changes

  • No training, by contract — commercial use runs under the Commercial Terms, where inputs and outputs aren't used for training. The toggle above is a consumer-only concern. [confirmed]
  • You're the data controller; Anthropic is the processor — admin controls over members, export of conversation history, and a Data Processing Addendum with Standard Contractual Clauses for cross-border transfer. [confirmed] (processor or controller?)
  • API: inputs and outputs auto-delete after ~30 days (sooner under a Zero-Data-Retention agreement, which stores nothing at rest), never used for training. [confirmed] (API data retention)

Where your data lives (matters under GDPR)

  • Direct use of Claude (claude.ai or the Anthropic API) is US-hosted — Anthropic doesn't offer an EU/UK data-residency option on the direct product. For most EA/AIS/nonprofit sharing this is fine; it bites only if you're under a residency obligation. [unclear] (no primary Anthropic page states a residency option — drawn from secondary coverage, WAIMAKERS GDPR guide, seen 2026-06-07)
  • Need EU residency? The route is running Claude through AWS Bedrock or Google Cloud Vertex AI in an EU region — a separate commercial setup, not a setting on claude.ai. [unclear] (same source)

Bottom line for sharing

A published Artifact is something you meant to hand out, so the training question barely applies to it. The data hygiene above is about your private workspace — and on a personal plan, one toggle (Help Improve Claude → off) settles it.

Sources