Claude data privacy depends on the plan: consumer claude.ai trains on your data by default (opt-out available); the API and business plans don’t. Retention, ZDR, HIPAA, and how to keep sensitive data out of Claude.
Claude’s privacy posture is genuinely strong for business use and weaker (by default) for consumer use. The deciding factor is the plan. Through the API, Claude for Work, and Enterprise, Anthropic doesn’t train on your data and offers enterprise controls like Zero Data Retention. On personal claude.ai plans, your conversations feed model training unless you opt out. If you only remember one thing: consumer and commercial Claude are different products for privacy purposes.
| Topic | Where Claude stands |
|---|---|
| Training on your data | Consumer (Free/Pro/Max): yes by default since Aug 2025, opt-out available. Commercial (API/Work/Enterprise): no. See does Claude train on your data. |
| Data retention | Consumer opt-out: 30 days. Consumer opted-in: up to 5 years. Commercial: per commercial terms; ZDR available. |
| Selling data | Anthropic states it does not sell your personal data. |
| Compliance | Anthropic reports SOC 2 Type II and ISO 27001; a BAA is available for HIPAA workloads on eligible plans. |
| Human review | Conversations may be reviewed for safety/abuse (e.g., flagged content), per Anthropic’s policies. |
These are Anthropic’s stated positions; always confirm the current terms for your plan, since consumer policy changed once already in 2025.
Most organizations focus on getting the enterprise agreement right — ZDR, no training, a BAA — and then assume they’re covered. They’re not, because employees still use personal Claude accounts for work, and those consumer accounts train on data by default. This is the shadow-AI gap: your policy protects the sanctioned path, but the unsanctioned path is where the sensitive data actually goes.
Strac gives you a control that doesn’t depend on which Claude plan an employee is on. Its browser DLP detects and redacts PII, PHI, secrets, and source code in real time — in the browser, before anything is sent to claude.ai, the Claude API, or Claude Code — with no proxy and no TLS interception. Whether the account trains on data or not, the sensitive part never leaves. And the same policy covers ChatGPT, Gemini, Copilot, and every other AI tool.
Comparing tools? See ChatGPT data privacy and Gemini data privacy, or Is Claude AI safe and Is Claude HIPAA compliant for safety and compliance specifics.
For business use, largely yes - the API and Claude for Work/Enterprise don’t train on your data, offer Zero Data Retention for qualifying use, and can support HIPAA via a BAA. For consumer plans, your data trains models by default (since Aug 2025) unless you opt out.
Anthropic states it does not sell your personal data. It may use consumer conversations to train models (unless you opt out) and may review flagged content for safety, but selling data is not part of its stated policy.
Yes. Under commercial terms, API traffic isn’t used for training, and Zero Data Retention is available for qualifying use cases - a materially stronger privacy posture than consumer claude.ai.
Anthropic reports SOC 2 Type II and ISO 27001, supports GDPR obligations, and offers a BAA for HIPAA workloads on eligible commercial plans. Consumer plans are not intended for regulated data. See our Claude HIPAA guide for detail.
Don’t rely solely on Anthropic’s settings - redact sensitive data before it reaches Claude. Strac does this in the browser across corporate and personal accounts, so PII, PHI, and secrets never leave in the first place.
Claude is privacy-strong for business use and weaker by default for consumer use — and the gap between them is where sensitive data leaks. The control that always holds is keeping that data out of Claude in the first place. Strac redacts it before it’s sent, on any plan. Book a demo to see it protect your AI usage.
.avif){kind=icon}
.gif)
.webp)
