🔒OpenAI Announces Private Safety Processing for ZDR Models
OpenAI's new safety system without breaking privacy vows
TL;DR
OpenAI launches Private Safety Processing, a mechanism to scan model interactions for risks while adhering to Zero Data Retention. Anthropic also relaxes its strict data retention policy slightly.
OpenAI has unveiled Private Safety Processing, designed to automatically detect safety issues in customer interactions with AI models without violating the Zero Data Retention (ZDR) commitment. This system evaluates not just prompts and outputs but also tool use and network signals for a more comprehensive risk assessment. For developers using ZDR-compatible models like Anthropic's Mythos 5 and Fable 5, this means enhanced safety measures while maintaining strict privacy controls. Anthropic now retains data for up to 30 days for covered models, allowing limited human review in specific scenarios.
Key Points
OpenAI's new system evaluates interactions across related prompts and outputs for better risk detection.
Anthropic retains data for up to 30 days for covered models like Mythos 5 and Fable 5, supporting safety work.
Human review by Anthropic personnel is limited to specific flagged content scenarios only.
OpenAI's Private Safety Processing ensures stored data remains inaccessible to company personnel under ZDR.
Apple, Google, Nvidia, and Meta offer similar private compute solutions for local AI model security.
Why It Matters
If you're using OpenAI or Anthropic models with Zero Data Retention policies, this update enhances safety without compromising privacy. For instance, developers working on sensitive projects can now benefit from advanced risk detection while adhering to strict data protection standards.
Frequently Asked Questions
Why does this matter?
If you're using OpenAI or Anthropic models with Zero Data Retention policies, this update enhances safety without compromising privacy. For instance, developers working on sensitive projects can now benefit from advanced risk detection while adhering to strict data protection standards.
What happened?
OpenAI launches Private Safety Processing, a mechanism to scan model interactions for risks while adhering to Zero Data Retention. Anthropic also relaxes its strict data retention policy slightly.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,264 builders reading daily.