AWS Open Source Agent Security Sandbox Rex: Enforces authentication at every step, thoroughly preventing AI from overstepping permissions

CryptoWorld News reports that AWS has announced the open-source release of Trusted Remote Execution (Rex), a policy-authorized script runtime. The tool is designed specifically for AI agents, and the core scripting language Rhai does not, by default, have any access permissions to physical hosts. Rex innovatively separates code execution from authorization policies entirely. If an agent’s script needs to perform system-level operations, it can only call the dedicated interfaces exposed by Rex. Before these calls reach the underlying system layer, they are submitted to the Cedar policy engine for forced evaluation. If an operation goes beyond the predefined boundaries, the agent will receive an access_denied_exception and will not cause actual damage. This mechanism enables enterprises to grant agents operational permissions such as reading logs or restarting services, while also setting strict security red lines through policies. Currently, Rex supports Linux and macOS environments, is developed in Rust, and is open-sourced under the Apache 2.0 license. The official guidance is to use it in conjunction with IAM and SSM.