As our sensitive data is increasingly carried over the Internet and stored remotely, security in communications becomes a fundamental requirement. Yet, today's security practices are designed around assumptions the validity of which is being challenged. In this thesis we design new security mechanisms for certain scenarios where traditional security assumptions do not hold. First, we design secret-agreement protocols for wireless networks, where the security of the secrets does not depend on assumptions about the computational limitations of adversaries. Our protocols leverage intrinsic characteristics of the wireless to enable nodes to agree on common pairwise secrets that are secure against computationally unconstrained adversaries. Through testbed and simulation experimentation, we show that it is feasible in practice to create thousands of secret bits per second. Second, we propose a traffic anonymization scheme for wireless networks. Our protocol aims in providing anonymity in a fashion similar to Tor - yet being resilient to computationally unbounded adversaries - by exploiting the security properties of our secret-agreement. Our analysis and simulation results indicate that our scheme can offer a level of anonymity comparable to the level of anonymity that Tor does. Third, we design a lightweight data encryption protocol for protecting against computationally powerful adversaries in wireless sensor networks. Our protocol aims in increasing the inherent weak security that network coding naturally offers, at a low extra overhead. Our extensive simulation results demonstrate the additional security benefits of our approach. Finally, we present a steganographic mechanism for secret message exchange over untrustworthy messaging service providers. Our scheme masks secret messages into innocuous texts, aiming in hiding the fact that secret message exchange is taking place. Our results indicate that our schemes succeeds in communicating hidden information at non-negligible rates.