OpenAI chases Anthropic's biz customers with zero data retention pledge

OpenAI has introduced Private Safety Processing, a mechanism designed to automatically scan customer model interactions for safety risks while adhering to Zero Data Retention commitments. This move could provide a competitive edge over Anthropic, which has yet to match this approach.

Under Private Safety Processing, customer model interactions are scanned for safety risks without violating Zero Data Retention. For commercial customers using ZDR, limited data retention and review will be implemented as part of the safety work. Prompts submitted to and outputs generated by covered models are retained for 30 days to support safety work on every platform.

The key distinction lies in how Anthropic and OpenAI define Zero Data Retention. Anthropic retains data for covered models, including Mythos 5 and Fable 5, for 30 days. In contrast, OpenAI allows customers to control infrastructure or choose to store data on OpenAI infrastructure, with OpenAI planning to provide customer-controlled encryption keys soon.

OpenAI's approach ensures that whatever data gets stored is unavailable to company personnel, except in cases of detected child exploitation material. This builds on existing automated protections and extends protections across related interactions, allowing automated systems to identify patterns without access to retained customer content.

While privacy concerns are paramount, the AI industry faces ongoing challenges in safeguarding sensitive data, particularly in light of the capabilities of AI agents. OpenAI's move to prioritize private model policing reflects a growing emphasis on protecting data flows from potential adversaries. However, experts caution that private inference alone may not offer comprehensive protection against all threats.