20 August 2026
OpenAI launches safety system that screens for misuse without storing data
First reported
Mindstream ran this on , a day before the other 2 sources picked it up.
- OpenAI built Private Safety Processing to detect harmful patterns across multiple conversations while keeping zero records of what users actually said.
- Enterprise customers can now use OpenAI's most advanced models with their data staying on their own servers or encrypted, with customers holding decryption keys.
- OpenAI says its safety system only receives narrow signals about misuse type and severity, not the actual content, unlike competitor Anthropic which stores data for 30 days.
Where they differ
The Rundown AI reported OpenAI's enterprise safety feature while Mindstream covered entirely different safety announcements for teenage ChatGPT users, including screen-time limits and parental alerts; only the first newsletter addressed the actual news about Private Safety Processing.
OpenAI launched Private Safety Processing, a system that scans for misuse across API sessions without staff viewing the content, maintaining its zero-data-retention promise.
OpenAI is introducing new controls for under-18s including voice reply toggles, quiet times, Study Mode scheduling, break reminders, and parental alerts for harmful prompts like those linked to eating disorders. The newsletter notes with mild irony that the chatbot is now enforcing screen-time boundaries.
Reported by The Decoder