Anthropic plans to let enterprise customers keep prompts and outputs on their own cloud while still requiring 30-day retention of business traffic.
The change, which Anthropic, the Claude chatbot maker, expects to deploy as a new safety system later this year, was developed in coordination with over 100 customers, including Salesforce, a source said.
The policy reverses part of an earlier approach after Anthropic had said in June it would apply 30-day retention to enterprise traffic on its Fable and Mythos models to guard against misuse.
The move also sits alongside rival product decisions, after OpenAI announced a safety system that avoids retaining customer data, highlighting diverging privacy trade-offs among major model providers.
The shift follows months of tightening retention rules and pressure from large corporate clients over compliance, governance and sensitive-data handling, and it signals Anthropic’s effort to give regulated customers more operational control without abandoning short-term safety monitoring.
Anthropic expects the new option to be available later this year as it phases enterprise traffic onto the updated safety stack.
