Whether Claude trains on your data depends on the Anthropic product, account type, and settings involved.
For companies, the most important distinction is between consumer Claude accounts and commercial environments such as Claude for Work, Anthropic API, Bedrock, Vertex, or Claude Code used under company-approved controls.
Anthropic's help center says that for Claude for Work, the customer organization controls submitted data and Anthropic acts as a processor. It also says Anthropic does not use Claude for Work customer data to train generative models.
That is a useful enterprise commitment, but it is not a substitute for employee training.
Employees need to know which Claude environment they are using
The phrase "I used Claude" is not specific enough for governance.
Training should teach employees to identify whether they are using a personal Claude account, Claude Team, Claude Enterprise, Claude Code, the Anthropic API, a cloud provider route, or a third-party tool that embeds Claude.
Each environment can have different controls, retention behavior, admin visibility, and approved-use rules.
That matters because an employee can make a risky decision even when the underlying model provider has strong enterprise protections. The risk often comes from using the wrong workspace, entering the wrong data, connecting the wrong files, or failing to review the output.
Claude Code creates a different training need
Claude Code is especially important for engineering teams because it can interact with local code context and developer workflows.
Anthropic's Claude Code data usage documentation separates consumer users from commercial users. For commercial users under Team, Enterprise, API, third-party platforms, and Claude Gov, Anthropic says it does not train generative models using code or prompts sent to Claude Code unless the customer has chosen to provide data for model improvement. The same documentation also explains operational behaviors around telemetry, error reporting, feedback, session quality surveys, and local session storage.
That means companies should give engineers workflow rules instead of stopping at "Claude Code is allowed" or "Claude Code is not allowed":
- which repositories can be used with Claude Code
- whether secrets, credentials, customer data, or regulated code areas are off limits
- how to review generated changes
- when to use pull request review instead of accepting suggestions directly
- which telemetry and reporting settings the company requires
- how Claude Code differs from browser-based Claude use
AI coding tools need developer-specific governance, not generic AI awareness.
The model-training question is only one part of the risk
"Does Claude train on your data?" is a good search query, but it is not the whole enterprise question.
Companies also need to decide:
- what data employees can enter
- what documents can be uploaded
- who can create shared projects or assistants
- whether external connectors are allowed
- how outputs should be checked
- when legal, security, or compliance teams need review
- how to handle model errors, outdated information, or missing context
A strong policy is useful only if employees understand how to apply it during work.
Projects, files, and shared workspaces need training
Claude adoption often expands from individual chat use into shared projects, uploaded documents, team knowledge, and developer workflows.
Those features can be useful, but they change the governance question. Employees need to know who can see shared materials, which documents are allowed, whether the workspace is approved for sensitive information, and how source material should be maintained.
For a commercial team, the issue is often not model training. It is whether employees are putting the right information in the right place, with the right audience, for the right workflow.
What to teach teams using Claude
Claude training should include approved workspace rules, sensitive data examples, source-grounded prompting, document handling, output review, escalation paths, and role-specific use cases.
Executives may need Claude for synthesis and decision support. Operations teams may use Claude for process documentation. Legal and compliance teams may need careful review protocols. Engineering teams may need Claude Code rules. Customer-facing teams may need strict source and approval requirements.
The same platform can require different training by role.
Questions leaders should settle before rollout
Before a company expands Claude usage, leaders should answer:
- which Claude products and workspaces are approved
- which teams can use Claude Code
- what information is too sensitive for Claude
- who can create shared projects
- how uploaded documents should be maintained
- what output review is required for customer, legal, engineering, or regulated work
- how employees should escalate uncertain cases
- how adoption and value will be measured
If those answers are not clear, employees will make inconsistent decisions.
What to teach Claude users
Claude users need a plain-language privacy checklist before they use the tool for work. Teach them to identify their account type, approved workspace, data rules, retention expectations, connector settings, and review requirements. The checklist should be specific to the company's Anthropic setup rather than copied from a generic training deck.
A useful exercise is to ask employees to classify five inputs: public website copy, an internal policy, a customer email, a contract excerpt, and a code file. For each one, they decide whether it can be used, whether it needs redaction, and who should approve it.
That exercise belongs in Claude training for teams, Anthropic Claude training, and AI governance training.
Practical takeaway
Anthropic says Claude for Work customer data is not used to train generative models. For commercial Claude Code use, Anthropic says code and prompts are not used for generative model training unless the customer has chosen to provide data for model improvement.
But companies still need training.
Employees need to know which Claude environment is approved, what information is allowed, how to review outputs, and when the workflow requires additional controls.
Safe Claude adoption depends on product policy and employee behavior.