OpenAI previews Private Safety Processing

OpenAI is trying to resolve the enterprise AI industry’s toughest tradeoff: stronger abuse detection without forcing customers to surrender sensitive data.

• –Automated systems analyze cross-interaction risk patterns while OpenAI receives only narrowly defined safety signals
• –Customer content can remain on customer-controlled infrastructure or be encrypted with customer-managed keys
• –The approach is especially relevant for healthcare, finance, cybersecurity, and agentic applications handling confidential workflows
• –ZDR remains subject to endpoint limitations and legal exceptions, including retention of images flagged for potential CSAM
• –The September white paper will be crucial for developers assessing whether the privacy claims hold up technically