to Create Smarter AI?
Amid fierce generative AI competition, Anthropic has revised its consumer terms of service and privacy policy for Claude. Free, Pro, and Max plan users can now choose whether to provide conversation data for model training. When consenting, data retention expands from 30 days to a maximum of 5 years. The tension between innovative AI advancement and personal information protection is again under scrutiny.
Effective August 29, 2025: users can opt in or out of having conversation and coding session data used for Claude model training; if consented, data is retained up to 5 years (deleting individual conversations removes them from training); applies only to Free, Pro, and Max individual plans (Claude for Work, Claude Gov, Claude for Education, API excluded); existing users must choose by September 28, new users decide during sign-up.
LLMs learn from vast data, but already-published internet text alone cannot reflect the latest code, new expressions, and real conversation contexts. This revision primarily intends to secure "field feedback" through real usage data — for example, debugging conversations are important training signals for solving similar problems. Anthropic competes intensely with OpenAI (ChatGPT), Google (Gemini), and Meta (LLaMA-based services); securing real usage data drives model quality improvements to maintain subscriber retention.
5-year retention can trigger sensitive privacy reactions, potentially conflicting with EU GDPR, Korea''s Personal Information Protection Act, and California CCPA. While Anthropic emphasizes data isn''t sold to third parties and undergoes filtering and anonymization, mere knowledge that an AI company retains personal conversations long-term creates unavoidable user anxiety. Among global comparisons, Anthropic''s 5-year retention period is the most aggressive. Core question: "How much personal data can we entrust to companies to create smarter AI?" Companies not securing data usage consent will fall behind in training quality; securing consent strengthens model performance but increases regulatory conflict risk. Regulators worldwide will likely intensively scrutinize AI company data retention and utilization policies.
