OpenAI Introduces New Privacy-Focused Safety Measures to Compete with Anthropic
As AI models grow more powerful and the risk of misuse increases, OpenAI has introduced a new privacy-centric safety system called Private Safety Processing. This automated monitoring tool is designed to detect potential abuse across multiple sessions without retaining any customer data. The move positions OpenAI against its rival Anthropic, which recently announced a policy allowing it to retain user session data for 30 days on certain models — a decision that has raised concerns among enterprise customers handling sensitive information. OpenAI’s approach builds upon its existing Zero Data Retention (ZDR) model by expanding monitoring capabilities beyond single sessions. The new system can analyze patterns across multiple interactions, helping detect malicious activities such as malware engineering spread over time. If triggered, it sends a narrowly defined signal to OpenAI for further action without human review of the data itself. Anthropic maintains ZDR in most cases but retains session data on specific models like Fable under its 30-day policy. The competition between these two AI labs is intensifying as both prepare for potential IPOs, with Anthropic reportedly having a $65 billion annualized revenue run rate and OpenAI trailing slightly behind.
Topics: AI safety Anthropic data privacy enterprise AI usage OpenAI