Yes on consumer plans (unless you opt out); no on commercial plans. This is the single most misreported fact about Claude, because the answer changed in 2025. If you use claude.ai on a Free, Pro, or Max account, Anthropic can now use your conversations and coding sessions to train its models unless you turn that setting off. If you use Claude through the API or a business plan, Anthropic does not train on your data. Knowing which bucket you’re in is the whole game.
Before 2025, Anthropic did not train on consumer conversations by default. That reversed with Anthropic’s updated consumer terms. Under the updated policy:
| Setting | What happens to your data |
|---|---|
| Stay opted in (default) | Chats and coding sessions may be used to train Claude; retention extends to up to 5 years |
| Opt out | Your new conversations are not used for training; retention stays at 30 days |
| Existing users | Were prompted to choose by September 28, 2025 |
The retention change is the part people miss: leaving the training setting on extends how long your conversations can sit in Anthropic’s pipeline from 30 days to as much as five years. Anthropic has said it de-identifies this data and does not sell it — but it is still your content, in a training set, for years.
Business use is treated very differently. Under Anthropic’s commercial terms, traffic through the API, Claude for Work, Enterprise, Education, and Government is not used to train models. Anthropic also offers Zero Data Retention (ZDR) for qualifying use cases, and can sign a BAA for HIPAA workloads. So a company building on the Claude API is in a fundamentally different position than an employee using a personal Claude Pro account — which is exactly where the risk hides.
Here’s the trap: your company may have an enterprise Claude agreement with no training and ZDR — but your employees also have personal Claude Pro accounts, and they paste work data into those. That consumer traffic is trained on by default. You don’t control Anthropic’s consumer policy, and you can’t see which account an employee used. The only durable control is to stop sensitive data from reaching Claude at all — on any account.
Strac’s browser DLP detects and redacts PII, PHI, secrets, and source code in real time — in the browser, before the text is submitted to claude.ai, the Claude API, or Claude Code. It works whether the employee is on a corporate or personal account, with no proxy and no TLS interception. So it doesn’t matter whether that particular Claude account trains on data or not: the sensitive part never leaves.
For the fuller picture, see Is Claude AI safe, Claude data privacy, Claude DLP, and Is Claude HIPAA compliant.
On consumer plans (claude.ai Free, Pro, Max) - yes, by default, since August 28, 2025, unless you opt out. On commercial products (API, Claude for Work, Enterprise, Education, Government) - no, your data isn’t used for training.
On a consumer plan, open your privacy settings and turn off the model-training / ‘help improve Claude’ option. Opting out also keeps retention at 30 days instead of up to 5 years. Commercial API and business accounts aren’t used for training in the first place.
If you leave the training setting on (consumer), retention extends to up to 5 years. If you opt out, it’s 30 days. Commercial products follow the commercial data terms, and Zero Data Retention is available for qualifying use.
No. Under Anthropic’s commercial terms, API traffic - along with Claude for Work, Enterprise, Education, and Government - is not used to train models.
Only if you control the account and the data. The safer approach is to redact sensitive data before it reaches Claude - because employees often use personal Claude accounts (which do train by default). Strac does this automatically in the browser.
Claude trains on consumer conversations by default now (since August 2025) and not on commercial API traffic. But you can’t police which account an employee uses — so the reliable control is to keep sensitive data out of Claude entirely. Strac redacts it before it’s ever submitted. Book a demo to see it in action.
.avif){kind=icon}
.gif)
.webp)
