FuseFSS: New Compiler Improves Speed and Efficiency of Secure LLM Inference
Researchers have developed FuseFSS, a compiler that streamlines secure two-server inference for large language models by replacing bespoke per-operator cryptographic protocols with a unified compilation pipeline. The system builds on function secret sharing (FSS), a cryptographic technique that allows clients to query hosted LLMs without exposing their prompts or internal data representations. FuseFSS achieves up to 1.50× end-to-end speedup and reduces online communication by up to 16% compared to prior state-of-the-art approaches, potentially making privacy-preserving AI inference more practical at scale.
FuseFSS is a compiler-based framework for two-server secure LLM inference, accepted at ICML 2026, that addresses a key bottleneck in existing function secret sharing (FSS) GPU systems: the inefficiency of handling fixed-point nonlinearities and helper operations through individually designed protocols. Instead of requiring bespoke cryptographic protocol design for each operator, FuseFSS introduces a unified pipeline where each scalar fixed-point operator is described by a compact specification covering its interval partition, low-degree arithmetic pieces, and required predicate bits. The compiler then emits two batched FSS evaluations — one packed comparison for predicate bits and one vector interval lookup for active coefficients — replacing the fragmented per-operator approach. Benchmarked on BERT and GPT-style models, FuseFSS delivers a 1.24×–1.50× end-to-end speedup and reduces online communication by 9%–16% without sacrificing model accuracy. Preprocessing overhead is also reduced, with key-generation time lowered by 14%–23% and key sizes shrunk by 20%–24%. These gains are significant because communication and preprocessing costs are primary practical barriers to deploying secure inference in real-world, latency-sensitive settings.
What's missing
The evaluation is limited to BERT and GPT-style models; generalizability to other architectures (e.g., mixture-of-experts or multimodal models) is unaddressed. The threat model assumptions — specifically what constitutes a semi-honest versus malicious adversary — and any security proofs or formal guarantees are not detailed in the abstract. Scalability to very large models (e.g., 70B+ parameters) is also not discussed.
What different sources said
- arXiv cs.AICenter
FuseFSS: Efficient Secure LLM Inference with Function Secret Sharing
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.