AI gets smarter, without watching you
TDT | Manama
Email: mail@newsofbahrain.com
OpenAI is testing a new system designed to make its AI models safer without requiring access to customers' private conversations.
Called Private Safety Processing, the system is aimed at business and API customers using OpenAI's most advanced models. It is designed to detect suspicious or harmful patterns across multiple interactions while maintaining Zero Data Retention protections.
The challenge is that increasingly capable AI agents can spread risky behaviour across several interactions, making it difficult to identify threats by examining a single request. OpenAI says its new approach can detect those patterns without giving its employees access to the underlying customer content.
When a potential risk is detected, OpenAI would receive only a limited signal about the type of activity rather than the actual prompts or responses. The system is currently being tested with selected customers, with a wider rollout planned later.
The move comes as businesses increasingly use AI for sensitive work, making privacy and security two of the biggest challenges in the next phase of AI adoption.
Related Posts