Anthropic has updated the data usage policy for its Claude Code tool, setting a 30-day limit on local session storage and confirming a ban on training models on the code of commercial users without their explicit consent.
What Happened
Local Claude Code sessions are now stored in plain text in the ~/.claude/projects/ directory for only 30 days, after which they are automatically cleared. For users on Team, Enterprise, and API plans, Anthropic guarantees that their code and prompts will not be used for model training without providing additional consent. Data retention for these categories is also limited to 30 days or governed by Zero Data Retention terms.
Context
This update aims to formalize the lifecycle of data that developers leave on their devices when using an agentic AI tool. This decision addresses transparency concerns regarding the use of proprietary code during the local development process.
Why It Matters for the Industry
For the industry, this is a step toward establishing security standards for AI development agents. A clear separation between local context and cloud training helps transform such tools from experimental solutions into reliable components of corporate CI/CD and Data Governance.
Why It Matters for Users
Developers need to account for the 30-day automatic cache deletion cycle when designing complex, long-lived tasks. Meanwhile, commercial users receive legal and technical guarantees protecting their intellectual property from being used in Anthropic's training sets.
Sources
Author
Look at AI, Editorial Staff