OpenAI Challenges Anthropic with New Privacy-First Safety Tools

The new service expands upon OpenAI's existing Zero Data Retention (ZDR) policy. While current ZDR protocols monitor abuse on a per-session basis, Private Safety Processing utilizes agents to analyze inputs and outputs across a longer horizon. By identifying patterns that might indicate complex threats—such as the creation of malware spread across multiple requests—the system flags potential misuse without requiring human review of private user interactions.

If the system detects suspicious activity, it transmits a narrowly defined signal to OpenAI rather than raw data. The company then determines whether enforcement is necessary, potentially contacting the customer to request additional context. This approach contrasts sharply with Anthropic’s "covered models" policy, which allows for human review of customer sessions through a controlled access path. Anthropic’s stance has faced pushback from enterprise clients concerned about the storage and inspection of sensitive data.

This competition over security protocols coincides with a broader rivalry between the two firms as they prepare for potential public offerings. With Anthropic’s annualized revenue reportedly reaching $65 billion and investor speculation regarding a $2 trillion valuation, OpenAI is under pressure to differentiate its enterprise offerings. By positioning its safety features as strictly non-retentive, OpenAI aims to capture customers hesitant to grant AI labs access to their proprietary information.