We explore a new technique for safe patch fingerprinting to automate vulnerability scanning of network servers. Our technique helps automate the discovery of inputs that safely discriminate vulnerable from patched servers for the latest vulnerabilities. This enables rapid updates to vulnerability scanning tools as new software vulnerabilities are discovered, allowing administrators to scan and secure their networks more quickly. To ensure such scans are safe and ethical, we need to reject inputs with malicious side effects. We have implemented a framework, based on delta execution, which tests the discriminative property of such inputs, as well as their safety. We use a fuzzer to find promising candidate inputs to further automate the process. To illustrate the potential of this approach, we present a Heartbleed case study.