OpenAI seeks to one-up Anthropic with new customer privacy protections
OpenAI has unveiled a new privacy-focused safety tool called Private Safety Processing, positioning it as a rival offering to Anthropic's recently announced data retention policy. The move comes as AI companies grapple with balancing enterprise customers' demands for privacy against the need to monitor for misuse of increasingly powerful models, and it marks an attempt by OpenAI to distinguish itself from Anthropic on this front.
Private Safety Processing extends OpenAI's existing Zero Data Retention approach by using an automated agent to analyse inputs and outputs across multiple sessions, rather than just one, without retaining customer data or requiring human review. This helps catch bad actors who might spread malicious requests, such as malware development, across separate conversations to avoid detection; if triggered, it sends OpenAI a "narrowly defined signal," after which the company may contact the customer, who can then choose whether to share further data. This contrasts with Anthropic's policy, announced in July, which allows it to retain conversation data for 30 days for "covered models" such as its Mythos-class systems, with human review possible only through a restricted, logged process — a policy that has unsettled some enterprise customers handling sensitive information.
- OpenAI launches Private Safety Processing, a privacy-first misuse monitoring tool.
- It analyses multiple sessions for abuse without retaining or exposing customer data.
- Positioned against Anthropic's 30-day data retention policy for "covered models."
New here? Start with this
OpenAI and Anthropic are two of the leading companies building advanced artificial intelligence systems, and both sell access to their AI models to businesses. As these systems become more capable, the companies face a tricky balance: business customers often want strong guarantees that their data won't be stored or read by anyone, while the AI firms also need some way to spot people misusing their tools for harmful purposes, such as writing malicious software.
Privacy has become a competitive selling point between the two firms. Anthropic set out its approach in July, allowing it to keep some conversation data for a limited period so it can be reviewed under certain safeguards if needed. OpenAI already offered a stricter option that avoids storing customer data at all, and has now built on it with new methods for detecting misuse without keeping or manually reading customer conversations.
This matters because large organisations, including those handling sensitive or confidential information, are increasingly relying on AI tools for everyday work, and how much control they have over their own data can influence which company they choose to work with. The rival approaches reflect a broader industry challenge: finding ways to prevent abuse of AI systems while still reassuring customers that their information stays private.