Over the past few months, we’ve worked with customers and partners to align safety policies across services that provide our models. We’re also updating our Usage Policies to make those requirements clearer and easier for developers to follow.

Models such as GPT‑6 Astra expand what developers and businesses can build. As those capabilities grow, we’re working to keep safeguards effective and consistently applied across the services that offer our models.

Before deploying a model, we evaluate its capabilities and risks, train it to follow safety requirements, and test whether its safeguards hold up when people deliberately try to bypass them. As we describe in our approach to safety and alignment, we combine safeguards designed to reinforce one another.

After deployment, we monitor for potential abuse and investigate concerns. To mitigate cyber misuse, for example, we train models to refuse harmful requests and use systems that look for suspicious patterns across interactions to decide how to respond. Trusted access programs provide another layer of protection and set eligibility requirements and conditions for access to our most powerful capabilities.

Some safeguards are built into the model and others depend on the systems and people around it. Both need to be effective whether people access our models directly or through a partner.

Developers and businesses should be able to choose where they access our models, with consistent safeguards across those services. We developed policies to align our safeguards across providers and are working with each of our partners to ensure their implementation meets or exceeds our safety standards.

Source link