Protecting sensitive data is no longer a problem restricted to governments whose national security is at stake. With ubiquitous Internet connectivity, it is challenging to secure a network -- not only to prevent attack, but also to ensure that sensitive data are not released. In this paper, we consider the problem of ensuring that only pre-authorized data leave a network boundary using either overt or covert channels, i.e., preventing exfiltration. We identify the goals of transparency, performance, and simplicity. A system designed to prevent exfiltration should not adversely affect the transfer of authorized data and should work with existing protocols. Key to our approach is: i) separating the process of vetting authorized objects from line-speed data verification; and ii) employing a restricted, but compliant, HTTP subset to limit covert channels. In our evaluation, we show that Glavlit adds little overhead to the operation of a software network bridge.