OpenAI has announced a new privacy-focused safety service, Private Safety Processing, designed to monitor for misuse of its AI models without retaining customer data. This move appears aimed at differentiating the company from rival Anthropic, whose recent data-retention policy has drawn criticism from some enterprise customers. The service, currently in preview with select customers, is an automated system that expands upon OpenAI‘s existing Zero Data Retention (ZDR) policy. While ZDR monitors for abuse on a per-session basis, Private Safety Processing can assess inputs and outputs across multiple conversations, allowing the detection of malicious activity that spans over time.

The system operates without human review of user conversations num conjunto. If triggered, it sends a ‘narrowly defined signal’ to OpenAI, warning of a specific type of activity. Based on that signal, OpenAI can decide whether enforcement is necessary. If so, the company will reach out to the customer for more context, and the customer may choose to share data at their discretion. This approach contrasts with Anthropic‘s policy for its ‘covered models,’ which includes retaining user conversations for 30 days to allow for safety analysis. Anthropic has stated that human review of customer data is possible, but only through a controlled access path with a small set of approved reviewers, and all review sessions are recorded in a tamper-proof log.

The announcement highlights the competitive tension between OpenAI and Anthropic. Both companies are seeking advantages, with recent reports indicating that OpenAI‘s Q2 growth was slower than Anthropic‘s, and Anthropic‘s annualized revenue run rate is reportedly $65 billion. Anthropic investors have suggested a potential IPO at $2 trillion, while OpenAI is also working on its IPO.