This framework, the Frontier Open-Weight AI Risk Management Framework, was developed jointly by Concordia AI and Z.AI and was written for open-weight developers and the wider ecosystem. It is a companion to the Frontier AI Risk Management Framework 2.0 and was first announced on 21 September 2026, at the UN Digital Cooperation Day.
Open-weight models advance transparency, research, and innovation, and spread the benefits of AI more widely. They also carry distinct risks. A release cannot be undone, safeguards can be removed through fine-tuning, and no single party can monitor use or withdraw the model. Managing these risks therefore requires an approach tailored to open release.
The Framework follows the six stages of risk management, from risk identification to risk governance. Many protocols in our earlier Framework 2.0 assume the developer controls deployment, for example by gating access or monitoring use. Open-weight releases shifts that control to the downstream ecosystem. At each stage, the Framework flags which protocols break down under open release and specifies an adapted approach. It also extends the analytical lens of Deployment Environment, Threat Source, and Enabling Capability with a fourth dimension, Societal Resilience, which captures how well society can absorb and respond to AI-enabled harms.
The Framework does not resolve every tension between openness and safety. It offers a first comprehensive, evidence-based foundation for navigating them. We invite researchers, developers, and policymakers to build on this work and advance the science and practice of open-weight AI governance.