OpenAI is previewing a new enterprise safety-monitoring system that it says can detect potential misuse without retaining customer data, according to TechCrunch. The service, called Private Safety Processing, is being offered first to select customers and is framed by the outlet as OpenAI’s answer to a growing enterprise AI problem: customers want strong privacy protections, while model providers want tools to detect harmful use. The tension is becoming more visible as frontier AI systems become more capable. TechCrunch reports that AI companies are trying to balance enterprise privacy commitments with safety guardrails designed to spot abuse, including activity that may be spread across multiple interactions rather than contained in a single prompt. OpenAI already offers Zero Data Retention, or ZDR, for some customers, according to TechCrunch. In that setup, agents inside the OpenAI application programming interface monitor for abuse on a per-session basis, while customer data is not retained by the company. Private Safety Processing is meant to widen that model. OpenAI describes it, according to TechCrunch, as long-horizon safety monitoring that can assess inputs and outputs across multiple conversations, not just one session. The aim is to catch patterns that a single-session screen may miss, while still avoiding human review of a user’s conversations. TechCrunch gives the example of a bad actor who might spread requests across separate sessions to avoid detection, such as someone hypothetically trying to engineer malware for a cyberattack. OpenAI told the outlet that Private Safety Processing can analyze those multiple conversations for signs of malicious use without retaining the underlying customer data. The enforcement mechanism is narrower than full data access, as described in the report. If the system is triggered, it may send OpenAI a “narrowly defined signal” warning of a specific type of activity. OpenAI can then decide whether enforcement is necessary, and may contact the customer for context or to work through the issue. TechCrunch reports that the customer may choose whether to share data with OpenAI. The move also lands against Anthropic’s recent retention policy. TechCrunch reports that Anthropic announced in July that it may retain user data — including sessions and conversations — for 30 days for “covered models.” The outlet says the policy was designed for safety analysis but has concerned some enterprises handling sensitive data. For OpenAI, the product pitch is specific: preserve a stronger privacy posture for enterprise customers while still monitoring for misuse that spans multiple conversations. For Anthropic, TechCrunch’s report suggests the competitive pressure is now partly about whether safety monitoring requires broader retention of customer sessions, or whether model providers can detect abuse through more limited signals. Who benefits: OpenAI benefits if enterprise customers view Private Safety Processing as a stronger privacy alternative to retention-based safety review. Customers with sensitive internal data may also benefit if the system works as described and limits what OpenAI receives during enforcement review. Who's exposed: Anthropic is exposed to customer pushback over its covered-model retention policy, according to TechCrunch. More broadly, AI vendors that rely on retaining customer conversations for safety review may face sharper enterprise scrutiny.