Secure computation lets mutually distrusting parties compute over private data while revealing nothing beyond the result, but a persistent gap separates the cryptographic techniques that make it efficient from the demands of deploying it against realistic, actively malicious adversaries. This proposal works at both ends of that gap. At the technique level, I develop a generic compiler that transforms any zero-knowledge (ZK) proof for SIMD circuits into one for general circuits while preserving efficiency, with concrete gains across several proof systems. At the system level, I show how privacy-preserving applications can be secured under active corruption: I enable an efficient way to fetch an encrypted entry from a secret database through a TEE-free, simulation-secure protocol tolerating actively corrupted adversaries. Together these results support a broader thesis: that advances in the cryptographic building blocks of secure computation can be systematically channeled into deployable systems. Looking ahead, the proposed work extends this methodology toward emerging multi-agent settings, where autonomous agents must compute over sensitive data without trusting one another.