The deployment of Fully Homomorphic Encryption (FHE) at scale is hindered due to its heavy computational overhead. While specialized hardware accelerators like Google Tensor Processing Units (TPUs) can help, mapping complex cryptographic kernels onto such architectures remains a challenge. Efficient execution requires co-optimization between the systolic array-based Matrix Multiplication Unit (MXU) and Vector Processing Units (VPUs), as well as the orchestration of data movement across the vector register files. Existing compiler stacks often abstract low-level hardware utilization, requiring developers to adopt a manual trial-and-error process that often results in fragmented execution and underutilized resources. To accelerate this development process, we use AlphaEvolve to automate the exploration of hardware-aware cryptographic-kernel optimizations. We frame optimization as an evolutionary search problem, utilizing the closed-loop system provided by AlphaEvolve, that leverages LLM-driven code generation. We use real-world feedback from hardware execution and rigorous correctness testing to guide the evolution process. We evaluate AlphaEvolve optimization on primitives for both the TFHE (Jaxite) and CKKS (CROSS) FHE schemes on Google Cloud TPUv5e, a contemporary TPU architecture. Within 24 hours of automated exploration, AlphaEvolve discovered implementation-level optimizations that improve TFHE bootstrap latency by 2.5x and CKKS rotation and multiplication latency by 1.31x and 1.18x, respectively, relative to human-engineered state of the art. These results demonstrate that AlphaEvolve can be used to enable researchers to navigate the optimization trade-offs between cryptography, compilers, and hardware accelerators.

In this work, we identify a set of side-channels in our Confidential Federated Compute platform that a hypothetical insider could exploit to circumvent differential privacy (DP) guarantees. We show how DP can mitigate two of the side-channels, one of which has been implemented in our open-source library.

Anonymous-credential schemes allow users to obtain credentials on various attributes, and then use those credentials to give unlinkable proofs about the values of some attributes without leaking anything about others. They have recently received interest from companies including Google, Apple, and Cloudflare, and are being actively evaluated both at the IETF and in the EU. Anonymous credentials based on BBS signatures are a leading candidate for standardization. In some natural applications of anonymous credentials, it is beneficial to hide even the issuer of a credential, beyond revealing the fact that the issuer is in some pre-determined set specified by a verifier. Sanders and Traoré recently showed a construction of such issuer-hiding anonymous credentials based on the Pointcheval–Sanders signature scheme. In this work we show how to achieve issuer hiding for BBS-based anonymous credentials. Our construction satisfies a notion of everlasting issuer-hiding anonymity, and is unforgeable in the generic group model. It can be integrated into existing standards, and has several efficiency advantages compared to prior work.

The stateless hash-based digital signature algorithm (SLH-DSA) is a post-quantum signature scheme based on the SPHINCS ^+ framework that was recently standardized by NIST. Although it offers many benefits, a drawback of SLH-DSA is that it has relatively large signatures. Several techniques have been proposed to reduce the signature size of SPHINCS-like schemes, and NIST is actively evaluating variants with shorter signatures for possible future standardization. We explore using forced pruning in the few-time signature scheme used by SPHINCS ^+ to reduce the overall signature size. Prior work suggested similar ideas, but claimed that the improvement from forced pruning was small. We re-visit this conclusion by performing a detailed theoretical analysis of forced pruning along with a more thorough exploration of its benefits. We show that forced pruning can improve upon SPHINCS+C (Oakland 2023) in all respects, and can reduce the overall signature size for the “smaller SPHINCS ^+ ” variants proposed by Fluhrer and Dang by up to 20

Succinct zero-knowledge arguments (zk-SNARKs) enable a prover to convince a verifier of the truth of a statement via a succinct and efficiently verifiable proof without revealing any additional information about the witness. A barrier to the practical deployment of zk-SNARKs is their high proving cost. With this motivation, we study server-aided zk-SNARKs, where a client/prover outsources most of its work to a single, untrusted server, while the server learns nothing about the witness, the statement, or even the final proof. We formalize this notion and show how to efficiently realize server-aided proving for widely deployed zk-SNARKs such as Nova, Groth16, and Plonk. The key building block underlying our designs is a new primitive, encrypted multi-scalar multiplication (EMSM), that enables private delegation of multi-scalar multiplications (MSMs). We construct an EMSM from variants of the LPN assumption in which the client does $O(1)$ group operations, while the server's work matches that of the plaintext MSM. We implement and evaluate our constructions. Compared to local proving, our techniques lower the client's computation by up to $18 \times$ and reduce the proving latency by up to $8 \times$.