Security & compliance
Anthropic data handling
What Anthropic receives for inference, and the contractual limits that govern how it is handled.
What we send
- Extraction: the text content of your contract documents, plus a schema specifying the obligations to extract.
- Analysis: the extracted obligations and the resolved playbook positions for comparison.
- Drafting and Ask: the slice of obligation data relevant to the prompt, plus the user’s question.
What Anthropic does with it
- Returns the inference response over the secure API.
- Retains data for up to 30 days for abuse and safety monitoring.
- Excludes customer API traffic from model training. This is committed in Anthropic’s API terms and our commercial agreement.
Models in use
- Extraction: Claude Sonnet, with a schema-locked tool definition so the output adheres to the obligation structure.
- Analysis, drafting and Ask: Claude Sonnet.
No training on your data
Customer API traffic is excluded from model training. This is a contractual commitment under Anthropic’s API terms and our commercial agreement.