OpenAI Launches Privacy-Centric Safety Tool to Challenge Anthropic

OpenAI has introduced a new service called Private Safety Processing, designed to monitor for AI misuse while ensuring no customer data is retained, as TechCrunch first reported. The automated system expands upon the existing Zero Data Retention (ZDR) policy by implementing "long-horizon safety monitoring." Unlike standard ZDR, which monitors on a per-session basis, this new technology analyzes inputs and outputs across multiple conversations to identify malicious actors who may spread requests over time to avoid detection, such as those attempting to engineer malware for cyberattacks.

If the system detects potential abuse, it sends a "narrowly defined signal" to OpenAI. The company then determines if enforcement is required and may contact the customer for further context; in such cases, customers can choose whether to share data at their own discretion, an OpenAI spokesperson told TechCrunch.

This move is seen as a strategic effort to outmaneuver Anthropic. In July, Anthropic announced a data-retention policy for "covered models," including Mythos-class models and Fable, which allows the lab to keep user sessions and conversations for 30 days to analyze potential impropriety. While Anthropic notes that human review occurs only through a controlled access path with a small set of approved reviewers and tamper-proof logs, the policy has caused concern among enterprises handling sensitive data.

The competition arrives as both firms eye potential IPOs. TechCrunch reports that Anthropic's annualized revenue run rate is approximately $65 billion, with some investors suggesting a potential $2 trillion IPO valuation, while OpenAI's Q2 growth was reportedly slower than that of its rival.