通过一致标准广泛获取

中文版正在随流水线自动整理，稍后刷新即可；可先阅读下方英文原文。

本文由 AI宝盒自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

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. These policies apply whether people access our models directly through OpenAI, partners that host them, or through reseller partners.

These policies require partners that provide access to OpenAI models to maintain safeguards consistent with our own, investigate potential violations, and take appropriate action when concerns are identified. That includes coordinating on enforcement so that restrictions remain effective across services providing access to our models.

Putting this into practice means accounting for how different services work and how customers use them. Our work with customers and partners on Private Safety Processing is one example of how we can develop effective safeguards while giving customers greater choice and control over their sensitive data.

We’re also updating our Usage Policies, effective November 20, 2026, to clarify the rules for safe model development and sensitive uses of AI. The updated Usage Policies build on that work by requiring services that provide direct access to our models such as model routers or AI gateways to get our review and approval before offering that access.

We’ll keep explaining what’s allowed, what’s restricted, and how customers and partners can ask questions or resolve compliance concerns.

As our models become more capable, we’ll keep improving how we detect, investigate, and respond to misuse. Customers can face misuse through their end users, compromised accounts, or leaked API keys and we work with them to understand and address those problems.

We’re also working with customers to introduce SafetyIDs, which customers can provide to help us distinguish between individual end users in an investigation without OpenAI learning those users’ identities. With SafetyIDs, we can respond to potential misuse more precisely while reducing disruption to others.

We base enforcement decisions on documented conduct and risk, apply our rules consistently, and respond in proportion to the circumstances. When we restrict access, we revisit those restrictions if meaningful changes in conduct, commitments, or safeguards address the concerns behind our decision.

We want more people to benefit from AI. Clear rules and consistent safeguards should make it easier for developers and businesses to build, compete, and choose the services that work best for them.

Auto-fetched and translated by AI Gem Box. All rights belong to the original author.