OpenAI seeks to one-up Anthropic with new customer privacy protections

OpenAI is previewing Private Safety Processing, an automated system for select customers that monitors potential abuse without retaining their data. The initiative positions OpenAI against Anthropic, whose July policy permits 30-day retention of user sessions and conversations for covered models, including all Mythos-class models and future models with similar capabilities. Anthropic introduced the policy to support safety analysis, but it has concerned enterprises handling sensitive information. Anthropic otherwise largely follows Zero Data Retention (ZDR), except for covered models such as Fable. OpenAI already uses ZDR agents within its API to monitor abuse on a per-session basis without retaining customer data or requiring human intervention. The company says Private Safety Processing expands ZDR into “long-horizon” monitoring: an agent assesses inputs and outputs across multiple conversations to identify misuse distributed over several sessions. For example, a bad actor could spread requests to engineer malware for a cyberattack and evade session-level detection. If triggered, the system may send OpenAI a narrowly defined signal identifying a specific type of activity. OpenAI can then determine whether enforcement is necessary. If so, it may contact the customer for context or cooperation; the customer retains discretion over whether to share data. The approach could strengthen OpenAI’s enterprise appeal by combining broader safety monitoring with privacy protections and no routine human review of conversations. Its commercial impact will depend on whether enterprises view cross-session detection and data non-retention as a meaningful advantage. Anthropic says human review can occur through a controlled access path involving a small group of approved reviewers. Each session is recorded in a tamper-proof log that reviewers cannot suppress or modify. The competitive stakes are significant. A recent report said OpenAI’s second-quarter growth was slower than Anthropic’s. Anthropic’s annualized revenue run rate is reportedly $65 billion; investors have said it could pursue an IPO at a $2 trillion valuation, while OpenAI is also working on an IPO.