Abstract. Decentralization is a major challenge for secure computing. In a decentralized setting, principals are free to distrust each other. The key challenge is to provide support for expressing and enforcing expressive decentralized policies. This paper focuses on declassification policies, i.e., policies for intended information release. We propose a decentralized language-independent framework for expressing what information can be released. The framework enables combination of data owned by different principals without compromising their respective security policies. A key feature is that information release is permitted only when the owners of the data agree on releasing it. We instantiate the framework for a simple imperative language to show how the decentralized declassification policies can be enforced by a runtime monitor and discuss a prototype that secures programs by inlining the monitor in the code.