Does Claude Code use my data for training?
By default, no. On consumer plans (Free, Pro, Max), Anthropic uses your Claude Code sessions for training only if you opt in. On commercial API, Team, and Enterprise accounts, your inputs and outputs are not used to train models. Exceptions on all plans: content flagged for safety review may be examined. Incognito chats are always excluded.
Why — the first-principles explanation
AI companies improve models by learning from data, so the real question is whose data and under what consent. Anthropic draws a hard line between consumer and commercial use, and Claude Code can run under either, which is why the answer depends on how you signed in.
On consumer plans (Free, Pro, Max), the default is that your chats and coding sessions are not used for training. You have to actively opt in—Anthropic asks, and you choose. If you say no (or ignore it), your work isn't fed into future models. On commercial accounts (the Anthropic API, Team, and Enterprise), the terms are stricter still: your inputs and outputs are not used to train models at all, because businesses need that guarantee to trust the tool with proprietary code.
Two caveats apply everywhere. First, safety: content flagged as potentially violating policy can be reviewed, and that review may inform safety systems. Second, retention isn't the same as training—Anthropic may store conversations for a period (up to five years for some data) for operational and safety reasons even when it isn't training on them. Data pulled in from connectors like Google Drive or MCP servers isn't used unless you copy it into Claude yourself.
An example that makes it click
Imagine a tutor who could get better by studying your homework. On the consumer plan, the tutor asks first: 'May I keep your worksheets to learn from?' If you don't say yes, they hand everything back and learn nothing from it. On the business plan, it's stricter—the tutor signs a contract promising never to study your worksheets at all. The only time anyone peeks is if a worksheet looks dangerous, like a note about setting a fire—that gets flagged for safety, not for teaching.
How to do it
- Check which account you're using: consumer (Free/Pro/Max) follows opt-in rules; API/Team/Enterprise are not used for training.
- On consumer plans, open Privacy settings in Claude and confirm the 'help improve Claude' training toggle is off if you don't want opt-in.
- For sensitive one-off work, use an Incognito chat, which is excluded from training regardless of settings.
- For guaranteed exclusion, use a commercial API, Team, or Enterprise account.
- Avoid pasting secrets you don't want stored; retention (up to five years for some data) is separate from training.
Key facts
- Consumer plans (Free, Pro, Max) use chats and coding sessions for training only if you opt in (default is off).
- Commercial accounts—Anthropic API, Team, and Enterprise—do not use your inputs or outputs for model training.
- Content flagged for safety review may be examined on any plan, regardless of training settings.
- Incognito chats are excluded from training use regardless of your settings.
- Anthropic may retain some conversation data for up to five years for operational and safety purposes—separate from training.
Anthropic's agentic coding tool that works in your terminal.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
Does Claude Code use your data to train its models? By default, no—but the details depend on your account. On consumer plans—Free, Pro, and Max—Anthropic only uses your chats and coding sessions for training if you actively opt in. If you don't say yes, your work isn't fed into future models. On commercial accounts—the API, Team, and Enterprise—it's stricter: your inputs and outputs are not used for training at all, which is exactly the guarantee businesses need before trusting it with private code. Two caveats apply everywhere. First, safety—content flagged as potentially harmful can be reviewed. Second, storage isn't training; Anthropic may keep some conversation data for a while for operational and safety reasons, even when it's not learning from it. Want extra protection for a sensitive task? Use an Incognito chat—it's excluded from training no matter what. Bottom line: opt-in on consumer, off-limits on commercial.
What authoritative sources say
People also ask
Does Claude Code train on my private code by default?
No. Consumer plans require opt-in; commercial API, Team, and Enterprise accounts are not used for training at all.
How do I stop my data being used for training?
On consumer plans, turn off the training toggle in Privacy settings, or use a commercial account or Incognito chat.
Is stored data the same as training data?
No. Anthropic may retain conversations (up to five years for some data) for operations and safety without training on them.
What about data from connected tools like Google Drive?
Connector data is excluded from training unless you manually copy that content into Claude yourself.