OpenAI is previewing a new service called Private Safety Processing, an automated system that watches for potential abuse while retaining none of the customer's data. The service is being offered first to select customers.
The move lands as a direct counter to rival Anthropic's recently announced data-retention policy, which allows the lab to keep user sessions and conversations for up to 30 days on "covered models" — including its Mythos-class systems and future models with similar capabilities — so potential abuse can be reviewed. That policy, announced in July, has unsettled some enterprise customers that handle sensitive data and resist having it stored or inspected by an AI lab.
OpenAI already adheres to a Zero Data Retention approach in its API, scanning sessions individually without retaining them. Private Safety Processing extends that model, promising misuse detection at scale without keeping any customer material.
As AI assistants absorb ever more sensitive workflows — legal drafts, medical summaries, financial data — the balance between surveillance and privacy is becoming a genuine enterprise procurement battleground, with each lab trying to turn safety monitoring into a competitive differentiator rather than a compliance burden.
