Safety
We will continue to offer Zero Data Retention for frontier models. As AI takes on longer, more autonomous work and delivers greater value to…
We will continue to offer Zero Data Retention for frontier models. As AI takes on longer, more autonomous work and delivers greater value to businesses, safety systems also need to identify risks acro
We will continue to offer Zero Data Retention for frontier models. As AI takes on longer, more autonomous work and delivers greater value to businesses, safety systems also need to identify risks across related interactions. To help address those risks, we're previewing Private Safety Processing, which is designed to improve safety without giving OpenAI personnel access to the underlying content.
Related
- Long-running models can solve hard open-ended problems, but their persistence can create safety risks that shorter-horizon evaluations miss.…
- As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforceme…
- Forecasting Trajectory-Level Safety Risks in Black-Box Multi-Turn Interactions
Source: OpenAI (X) | 2026-08-19