OpenAI Debuts Privacy-First Safety Tool to Counter Anthropic

OpenAI is taking direct aim at Anthropic on the enterprise privacy front. The company is previewing a new service it calls Private Safety Processing, an automated monitoring system designed to watch for potential misuse while retaining none of a customer's data.

What happened

The announcement, reported by TechCrunch, positions OpenAI squarely against Anthropic's recently expanded data-retention policy. In July, Anthropic said it would keep user sessions and conversations for 30 days for "covered models", its Mythos-class and similar future systems, so the lab could analyze them for potential impropriety. That move has worried enterprises that handle sensitive data and don't want it stored or inspected by the AI lab.

OpenAI's counter is built on Zero Data Retention, the API policy most AI labs already follow. Private Safety Processing widens ZDR's scope from a single session to long-horizon monitoring: an agent assesses inputs and outputs across multiple conversations at once, catching bad actors who spread malicious requests across sessions to avoid detection, all without any human review of user conversations.

Why it matters

For AI builders and enterprise teams, this is a real competitive signal. Data privacy and retention terms are becoming a deciding factor in which foundation model a company standardizes on. OpenAI is betting that "we don't keep your data, but we still catch abuse" is the stronger pitch than Anthropic's "we keep it briefly, and only inspect through a controlled, logged path."

The stakes are high on both sides. Anthropic's annualized revenue run rate is now reportedly $65 billion, and both labs are racing toward IPOs. Privacy posture, not just model performance, is now part of that race.

When Private Safety Processing triggers, OpenAI says it sends only a "narrowly defined signal" flagging a specific type of activity. The company then decides whether enforcement is warranted and can reach out to the customer for context, with data shared only at the customer's discretion. The service is in preview for select customers for now, but if it lands broadly, data-retention guarantees are likely to become a core part of enterprise AI procurement and a fresh battleground between the two leading labs.