OpenAI Reaffirms Zero Data Retention, Previews "Private Safety Processing"

OpenAI has reaffirmed its Zero Data Retention (ZDR) policy for eligible API customers and previewed a new companion system called Private Safety Processing, designed to let OpenAI detect misuse patterns across related interactions without giving staff access to the underlying content.

Details

• Zero Data Retention, core promise: OpenAI does not retain prompts or model responses after a request is processed, customer content is unavailable to OpenAI personnel for review, and enterprise data is not used for model training without explicit opt-in
• Eligibility: The announcement describes ZDR as available to “eligible API customers” using “frontier models,” but does not specify which particular models or exact eligibility criteria beyond that
• The problem it addresses: OpenAI says prior frontier-model deployments required customers to permit some content retention to enable safety monitoring, which could conflict with customers’ own security obligations — Private Safety Processing is meant to resolve that tension
• How Private Safety Processing works: it analyzes customer content across related interactions to spot misuse patterns, operates regardless of whether content is stored on customer-controlled infrastructure or OpenAI’s systems, uses customer-controlled encryption keys for anything stored on OpenAI systems, and returns only a “narrowly defined signal” about the type of activity to OpenAI staff — personnel cannot see the flagged content itself, even when a risk is identified; customers can appeal using their own system information and choose whether to share further details with OpenAI
• Timeline: Private Safety Processing is currently in testing with early customers, with a broader rollout and a technical white paper planned for September
• Customer quote: Glean CISO Sunil Agrawal is quoted saying enterprise AI adoption “depends solely on customer control of data” and that the approach “shows safety can advance without compromising…privacy”

What happened next

The announcement is light on specifics that would let enterprise buyers evaluate exact scope — which frontier models qualify, and what the formal eligibility bar is — leaving those details for the promised September white paper. The core tension it’s trying to resolve (safety monitoring vs. strict no-retention guarantees) is a recurring theme for enterprise AI vendors, and Private Safety Processing represents OpenAI’s specific technical answer: pattern-level signal without content access, similar in spirit to techniques increasingly used for content moderation on encrypted platforms elsewhere.