Skip to content

Back to the catalog

privacy.claude.comarticle

Is my data used for model training?

Anthropic's privacy help article on whether customer data trains its models. For commercial products (Claude for Work, Anthropic API, Claude Gov), inputs and outputs are not used for training by default. Thumbs up/down feedback stores the whole conversation for up to 5 years, de-linked from user IDs, and may be used for training. Team and Enterprise owners can disable feedback in organization settings.

TL;DR Explains that Anthropic does not train on commercial-product data by default, except for submitted feedback, which is kept up to 5 years.
Document

What it claims

Anthropic's help article states that inputs and outputs from its commercial products (Claude for Work, the Anthropic API, Claude Gov) are not used for model training by default. The article covers only commercial plans. Consumer plans (Claude Free, Pro, Max, and Claude Code used from those accounts) fall under a separate article with its own terms. Training use on commercial plans happens in two cases. The first is when a user explicitly sends feedback or a bug report, for example with the thumbs up/down button. The second is when a customer otherwise chooses to allow it. Either case can cover both chats and coding sessions.

How it works

A thumbs up/down rating stores the entire related conversation for up to 5 years. The stored record includes message content, custom styles, conversation preferences and model settings. Raw content from connectors such as Google Drive, including remote and local MCP servers, is excluded. Connector content is included only if it was copied directly into the conversation.

Anthropic says it de-links feedback from user and customer IDs before using it. It also says it does not combine feedback with the user's other conversations. Stated uses go beyond training: analysing service effectiveness, research and studying user behaviour, "as permitted under applicable laws."

flowchart TD
    A[Account type] -->|Free, Pro, Max| B[Consumer policy, separate article]
    A -->|Work, API, Gov| C{User sends feedback or bug report?}
    C -->|No| D[Not used for training]
    C -->|Yes| E[Whole conversation stored, de-linked]
    E --> F[Training, research, behaviour analysis]
    G[Owner disables Rate chats] -.blocks.-> C

How to use it

On Team and Enterprise plans, a Primary Owner or Owner can turn off thumbs up/down submission for all members. The control is the "Rate chats" setting under Organization settings > Data and Privacy. The page names no equivalent control for API-only customers. It points to the Trust Center and Commercial Terms for further detail and privacy contacts.

Limits

The page does not define what "otherwise choose to allow us to use your data" covers. It does not say whether a stored feedback record can be deleted before 5 years. It gives no retention terms for commercial data that is not submitted as feedback.

A developer using Claude Code under a personal Pro or Max account is governed by the consumer policy, not this one. Pasting connector or MCP output into a chat brings that content inside the feedback record if the chat is later rated.

Organisations named here

  • Anthropic
  • Claude Code

Technologies named here

  • Anthropic API
  • Claude for Work
  • Claude Gov
  • Claude Code
  • MCP servers
  • Google Drive
  • Claude
  • Model Context Protocol