Enterprise customers gain a new option to detect coordinated misuse across sessions while avoiding the data-holding practices that have alarmed some buyers. OpenAI said it is previewing Private Safety Processing to a subset of customers, an automated agent-based system that watches for potential abuse yet “retains none of the customer’s data.” The company positions the feature as an expansion of existing zero data retention tooling, applying monitoring across multiple conversations rather than limiting scrutiny to a single session.
The launch directly contrasts with Anthropic’s data-retention rule announced in July, which allows the lab to keep user sessions and their contents for 30 days for so-called “covered models.” Anthropic lists Mythos-class models and “future models with similar capabilities” as examples of models subject to that policy. The lab framed the retention as a safety measure, enabling analysis of suspicious activity, but several enterprise customers that handle sensitive material have pushed back at the idea of their conversations being stored and inspected.
OpenAI says Private Safety Processing detects patterns that unfold across sessions, such as fragmented requests intended to evade session-level checks. A spokesperson told TechCrunch the system can flag malicious behavior without human review of customers’ conversations. If the agent is triggered, it may send a “narrowly defined signal” to OpenAI indicating a specific kind of activity, and the company can then determine whether “enforcement is necessary.” In cases where enforcement proceeds, OpenAI says it will reach out for context, and customers may choose to share more data at their discretion.
Anthropic has defended its approach, noting that human review of customer data is possible only “through a controlled access path” that involves “a small set of approved reviewers,” and that every review is “recorded in a tamper-proof log that reviewers cannot suppress or modify,” the company says. That audit trail aims to reassure customers, but the retention window itself remains the core concern for firms with strict data governance requirements.
The timing sharpens competition. A recent report found OpenAI’s Q2 growth lagged Anthropic’s, and Anthropic’s annualized revenue run rate is now reportedly $65 billion. Some of Anthropic’s investors have suggested it could go public at $2 trillion, while OpenAI is also working toward an IPO. The privacy and safety features companies choose will influence enterprise purchasing, and both labs now have an incentive to press advantages in procurement conversations as they move toward public markets.
