Claude safety profile

What it does with your data

Claude follows two sets of rules. Free, Pro and Max are consumer plans. Team, Enterprise and the API fall under Anthropic’s Commercial Terms.

On consumer plans, Anthropic trains new models on chats and coding sessions only when the model training setting is on. New users pick this during signup, and the pages checked do not say which option is preselected. Even with the setting off, chats can still be used to improve models in three cases: conversations flagged for safety review, anything sent through thumbs up or thumbs down feedback, and programs you opt into, such as the Trusted Tester Program. Before chats are used for training, they are separated from the user ID, such as an email address.

Retention follows the same choice. With training allowed, chats are kept for 5 years. With training off, they are kept for 30 days. Feedback is kept for 5 years.

On commercial plans, the consumer training rules do not apply, but the pages checked do not state the commercial training policy directly.

Settings to change first

If anyone at your organization uses a personal Free, Pro or Max account, check these before they paste in work material:

1. Open “Privacy Settings” and turn off the model training setting. You can change it at any time.
2. Delete chats you no longer need. Deleted chats and deleted accounts are left out of future training.
3. Skip the thumbs up and thumbs down buttons on any chat with sensitive content. Feedback can be used for training and is kept for 5 years.

Turning the setting off stops new chats from being used and drops stored chats from future training runs. It does not remove data from training runs that already started or from models already trained, so change it on the first day.

On Team and Enterprise, the vendor pages checked do not describe a training setting or a menu path for admins.

Free and paid tiers

Free, Pro and Max are handled the same way under the Consumer Terms. Training depends on the user’s setting, and retention is 5 years or 30 days depending on that choice. The pages checked do not separate Free from Pro or Max, so paying for Pro does not by itself change how data is handled.

Team, Enterprise and the API have their own rules:

• API inputs and outputs are deleted within 30 days, except for features you control (such as the Files API), zero data retention agreements, Usage Policy enforcement and legal requirements.
• On Claude for Work and Enterprise, chats stay in the product until deleted. A deleted chat leaves history right away and backend storage within 30 days.
• Chats flagged for Usage Policy violations are kept up to 2 years.

On commercial plans, your organization is the controller of the data.

Admin controls

Developer Platform users and Enterprise Plan users on usage-based billing can choose a traffic routing location. By default, traffic may go to select countries in the US, Europe, Asia and Australia, and commercial data is stored in the US. Anthropic links to articles on custom data retention controls for Enterprise and on Incognito chats for organizations, but the pages checked do not describe either. They also do not cover single sign-on, user management or audit logs.

For certifications, the API and Enterprise are listed with SOC 2 Type 2, ISO 27001, ISO 42001, CSA Star, HIPAA and NIST 800-171. Team has the same list, with HIPAA marked N/A. Free, Pro and Max do not appear in the certification table.

Using it with donor and client data

Put work on Team or Enterprise, not on personal Free, Pro or Max accounts. Commercial plans are outside the consumer training rules, your organization is the controller, and they appear in the certification list. Personal accounts depend on each staff member getting one setting right.

Whatever the plan, set a clear rule about what never goes into a chat:

• Donor names next to gift amounts
• Client case notes
• Health details
• Payment or bank details

Use made-up names or general descriptions instead. For example, ask for help with “a thank-you letter to a monthly donor” rather than pasting a real donor record.

Before staff use Claude with donor data, get these points in writing from Anthropic, because the pages checked do not cover them:

• Whether Team or Enterprise chats are used for training
• What single sign-on, user management and audit log options exist
• How the Enterprise retention controls work

Questions people ask

Does Anthropic train on our chats?

On Free, Pro and Max, only when the model training setting is on, plus flagged chats, feedback and opt-in programs. On Team and Enterprise, the consumer rules do not apply, and the pages checked do not state the commercial policy directly.

If I turn training off today, is my old data removed?

Stored chats are dropped from future training runs. Data already in training runs that started, or in models already trained, stays included.

Where is our data kept?

For commercial products, Anthropic says data is stored in the US and may be processed in select countries in the US, Europe, Asia and Australia. The pages checked do not state a region for Free, Pro or Max.