Anthropic changes data retention policy for enterprise customers

On August 19, 2026, OpenAI announced Private Safety Processing, a new automated safety monitoring system that analyzes multiple interactions without retaining customer data. This launch directly targets rival Anthropic, whose July 2026 policy of retaining data from advanced models for 30 days drew backlash from enterprise customers. In response, Anthropic plans to alter its policy later in 2026, allowing businesses to host the required 30-day data retention on their own cloud infrastructure.

OpenAI Private Safety Processing launch

• ▪OpenAI's Private Safety Processing system is designed to detect patterns and safety risks across multiple interactions with its models without retaining customer data
• ▪When OpenAI's Private Safety Processing system is triggered, it sends a narrowly defined signal to OpenAI, allowing the company to decide if enforcement is necessary
• ▪OpenAI announced on August 19, 2026, that it is previewing a new automated safety monitoring system called Private Safety Processing to select customers

Anthropic data retention policy controversy

• ▪Anthropic's July 2026 data retention policy allows human review of customer data by a small set of approved reviewers, with each session recorded in a tamper-proof log
• ▪Anthropic plans to roll out a safety system later in 2026 that allows business customers to retain their data for 30 days on their own cloud infrastructure instead of Anthropic's
• ▪Anthropic announced a policy in July 2026 that allows the company to retain user data from "covered models," including Mythos-class models, for 30 days for safety analysis

AI safety monitoring approaches

• ▪OpenAI and other AI companies use Zero Data Retention policies, which employ agents within the API to monitor for abuse on a per-session basis without retaining data
• ▪OpenAI's Private Safety Processing expands on Zero Data Retention by using an agent to analyze inputs and outputs across multiple conversations to detect malicious activity

Enterprise customer privacy concerns

• ▪AI companies face a challenge in balancing the privacy demands of enterprise customers with the need to monitor AI usage for potential abuse and safety risks
• ▪Enterprise customers handling large amounts of sensitive data expressed concern and aggravation over Anthropic's July 2026 policy of harboring and inspecting user data

4 sources