Java bytecode verification for secure information flow

Security of Java programs is important as they can be executed in different platforms. This paper addresses the problem of secure information flow for Java bytecode. In information flow analysis one wishes to check if high security data can ever propagate to low security observers. We propose a static analysis similar to the type-level abstract interpretation used for standard bytecode verification. Instead of types, our technique works with secrecy levels assigned to classes, methods' parameters and returned values, and handles implicit information flows. A verification tool based on the proposed technique is under development. Using the tool, programs downloaded from untrusted hosts can be checked locally prior to executing them.