Anthropic API Usage Policy

Anthropic's Acceptable Use Policy sets out what users across all product surfaces (Claude.ai, API, cloud integrations) may and may not do with Claude. The policy's most operationally significant provisions establish mandatory qualified human review and AI disclosure requirements before AI-generated recommendations are delivered in high-stakes domains including medical diagnosis, legal advice, credit decisions, employment screening, insurance underwriting, and housing approvals. Additionally, the policy prohibits using model outputs to train other AI models without Anthropic's prior authorization, directly affecting developers who may seek to use API outputs for fine-tuning or distillation purposes.

This document is Anthropic's Acceptable Use Policy (AUP), effective November 12, 2026, governing all users who submit inputs to Anthropic's products and services, including Claude.ai, Claude Code, API consumers, cloud provider customers, and end users of third-party integrations. The agreement establishes three tiered categories of restrictions: Universal Usage Standards (absolute prohibitions applicable to all users), High-risk Use Case Requirements (mandating qualified human-in-the-loop review, AI disclosure, and independent safety controls for AI recommendations in legal, medical, financial, credit, insurance, housing, employment, education, healthcare access, public benefits, legal status, and physical action domains), and Additional Use Case Guidelines covering consumer-facing chatbots, products serving minors, agentic deployments, and directory-listed software. Notably, the policy reserves enforcement authority to Anthropic's Safeguards Team, which may warn, throttle, limit, suspend, or terminate access upon suspected violations, and it explicitly prohibits use of model outputs to train competing AI models without prior authorization, a provision with direct implications for developers and enterprise API consumers. The policy engages frameworks including GDPR, CCPA, COPPA, the EU AI Act, FTC Act, HIPAA, and sector-specific regulations governing employment, credit, housing, and financial services, with several High-risk AI Recommendation categories corresponding directly to regulated activities under these frameworks. Compliance obligations vary significantly by jurisdiction, and the policy itself acknowledges that local laws take precedence where they conflict with or add requirements beyond those stated in the AUP.

The agreement establishes that individuals receiving advice or subject to decisions based on High-risk AI Recommendations must be clearly informed that AI was involved in producing the recommendation. Under these terms, operators deploying Claude in high-risk domains are required to maintain qualified human oversight before AI-generated recommendations reach affected individuals, which creates a procedural layer between model output and consequential decisions in areas such as healthcare, credit, employment, and housing. You can report potentially inaccurate, biased, or harmful model outputs by contacting usersafety@anthropic.com or using the in-product feedback mechanism where available.