Ladon: Strengthening Confidential Computing by Decentralizing Trust in the Key Management Service via a Threshold KEM
Abstract
The Key Management Service (KMS) plays a central role in confidential computing: it wraps the symmetric keys used to encrypt asset owners’ sensitive data (e.g., healthcare records, proprietary models), and later releases those keys to a trusted execution environment running the intended workload. Despite this critical role, current KMS deployments remain vulnerable to a single point of compromise: a breach of the KMS can expose the asset keys and, consequently, every asset they protect. To avoid this single point of compromise, a Post-Quantum (PQ) secure threshold Key Encapsulation Mechanism (KEM) can replace the traditional KMS, wherein the asset owner encapsulates the asset key under a single public key, while decapsulation is distributed across a committee of KMS servers. Unfortunately, existing PQ-secure threshold KEMs impose high communication overhead during asset-key decapsulation and produce large public keys and ciphertexts.
In this work, we present Ladon, an efficient PQ-secure threshold KEM that addresses the limitations of prior work in communication during threshold decapsulation, public-key, and ciphertext sizes. Ladon builds on NIST’s Module-Lattice Key Encapsulation Mechanism (ML-KEM) standard, whose underlying design principles enable more efficient parameter selection than other designs. These yield three prominent advantages: (i) Ladon’s encapsulation and decapsulation are functionally equivalent to ML-KEM, inheriting its well-established security guarantees; (ii) Ladon’s threshold protocol incurs lower communication overhead than the existing PQ-secure threshold KEMs, along with up to 8x smaller public keys and 5x smaller ciphertexts; and (iii) with Ladon, the asset owner encapsulates only once, regardless of the KMS committee size, making the threshold structure transparent to them. We also open-source a Rust implementation of Ladon.
