OpenAI unveils privacy-focused monitoring tool to edge out rival Anthropic

OpenAI is testing a new service called Private Safety Processing with select customers, aiming to detect AI misuse without holding onto user data. The automated system reviews inputs and outputs across multiple conversations, rather than just a single session, to spot patterns that might indicate abuse over time.

The move appears to counter Anthropic's recent policy of retaining session data for 30 days on certain models, which has frustrated some businesses handling sensitive information. OpenAI's existing Zero Data Retention approach already avoids storing customer content, but the new system widens that scope by analyzing long-term behavior across sessions.

If the system flags suspicious activity, it sends a narrow signal to OpenAI, which can then decide whether to contact the customer for more context. Human review is not part of the process unless the customer chooses to share data voluntarily.

Anthropic, meanwhile, says any human review of customer data happens only through a controlled path with approved reviewers and tamper-proof logs. The competition between the two firms is intense, with recent reports showing OpenAI's growth slowing compared to Anthropic, whose annualized revenue run rate is reportedly around $65 billion and which investors say could go public at a $2 trillion valuation.

Source: TechCrunch