OpenAI is testing a privacy-focused safety system that can detect misuse across multiple interactions without retaining customer data, contrasting with Anthropic’s new 30-day retention policy.

OpenAI is reviewing a new technique to safely serve its most advanced models to businesses without needing to retain any of their data fragments. The latest update comes as Anthropic has announced that it would institute a 30-day retention policy for business customers who want to use Fable 5 and Mythos 5.

OpenAI is testing “Private Safety Processing” with early customers. The system is designed to identify misuse patterns across related interactions while preserving zero data retention, or ZDR, security. As per a new update, OpenAI says its systems would send the company a narrowly defined safety signal without exposing the underlying prompts or responses.

Customer data would stay on customer-controlled infrastructure or be stored by OpenAI with encryption keys controlled by the customer. The company is also planning a broader rollout and a technical white paper in September.

OpenAI and Anthropic on opposite ends

Aleah Houze, Head of Product Policy at OpenAI, said in a briefing that, “We’re seeing with more capable frontier models that often risks are emerging not just by looking at one single prompt and response pair, but when you look over time at multiple interactions.”

techMore from Tech

She further added, “So, an example of this would be somebody might be asking about the weakness in a company’s software in one conversation, and then later in another conversation they might ask about remote access or what security tools can detect.”

While OpenAI is planning to bring in an update that would not require user data retention, Anthropic says that retaining data is necessary for security.

Anthropic recently said in a risk report, “We have recently announced our plan to require 30-day data retention on our most capable models—a decision we believe will be unpopular with customers who have come to expect zero retention, and pose real risks to our business success (especially if competitors do not follow), but which we believe is essential to detect and prevent sophisticated attacks that span multiple requests.”

The new safety systems are currently being curated for eligible enterprise and API customers, not people using OpenAI’s paid or subscription consumer ChatGPT plans.

OpenAI’s ZDR controls do not apply to Free, Plus, Go and Pro users, and their existing consumer data settings remain unchanged.