OpenAI Unveils Private Safety Processing to Outpace Anthropic on Privacy
OpenAI is previewing a new privacy-focused safety service called Private Safety Processing, aiming to one-up rival Anthropic on customer privacy. The automated system monitors for potential abuse of AI models while retaining none of the customer's data. The move contrasts sharply with Anthropic's recently announced policy of retaining user session data for 30 days for certain covered models, a stance that has frustrated some enterprise customers.
Anthropic's data-retention policy, announced in July, applies to its Mythos-class models and future models with similar capabilities. The company says the policy exists for safety purposes, allowing it to sift through and analyze potential misuse. However, enterprises handling large amounts of sensitive data have expressed deep concern about having their information stored and inspected by the AI lab. Notably, Anthropic otherwise largely adheres to a Zero Data Retention (ZDR) approach, except for these covered models.
OpenAI says Private Safety Processing extends the scope of ZDR through long-horizon safety monitoring that assesses inputs and outputs across multiple conversations rather than just one. An agent conducts the monitoring and, if triggered, analyzes interactions across sessions for signs of misuse. According to a spokesperson, the technology helps OpenAI detect malicious activity — such as a bad actor spreading out requests to engineer malware — that unfolds over multiple sessions, all without retaining customer data or requiring human intervention.