Transforming animals in a cyber-behavioral biometric menagerie with Frog-Boiling attacks

While recent research has demonstrated how frequent updating of users' templates can enhance the performance of a biometric system, there has not been much work devoted to studying the effects of attacks against template update mechanisms. In this work, we present an attack which stealthily leverages the template update scheme of a keystroke verification system to poison users' templates. Using a publicly accessible dataset and some of the best performing individual and fusion verifiers in keystroke authentication, we show how the attack increases the error rates of the verifiers as it transforms groups of well performing users into ill performing users. In our experiments, depending on the template towards which the attack is made to converge, equal error rates of verifiers increased from between 9.9% and 18.9% to between 19.1% and 63.6% as a result of the attack. Results demonstrated in this paper call for research on new biometric sample attestation and validation techniques to augment template update mechanisms.