Updates to Consumer Terms and Privacy Policy

Source: Anthropic Date: 2025-08-28 URL: https://www.anthropic.com/news/updates-to-our-consumer-terms

Summary

Anthropic updated consumer terms effective October 8, 2025. Key change: opt-in permission for users to allow conversation data use for model training and safety systems. Data retention: extends from 30 days to 5 years for users who opt in to training; 30-day retention unchanged for those who decline. Scope exclusion: updates do not apply to commercial services (Claude for Work, Government, Education, API/Bedrock/Vertex). Users can adjust preferences via Privacy Settings anytime. Rationale: AI development cycles require 18-24 months of data; opt-in framing preserves user control.

Implications

• Data/privacy / opt-in training data collection. Requiring opt-in (not defaulting to opt-out) for training data use is the GDPR-compatible approach — it’s stronger than the minimum required by most privacy frameworks. The opt-in design reflects European privacy expectations applied globally.
• 5-year retention for training users. Five years of conversation history for training opt-in users is a long retention period — it spans multiple model generations. This is the data asset that powers future model improvements without requiring new data collection infrastructure.
• B2C exclusion of commercial services. The explicit exclusion of Work, Government, Education, and API users is the enterprise data isolation guarantee — enterprise customers’ conversations will never be used for model training regardless of consumer policy changes.
• October 8 deadline for existing users. The 40-day notice period (August 28 update → October 8 effective) is the minimum for a major policy change — it gives users time to make an informed decision but creates urgency.
• Watch: what fraction of users opted in vs. declined; whether the 5-year retention window was challenged by European data protection authorities; how the training data opt-in affected model improvement trajectory.