Entity trail

Homomorphic Encryption

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-15 / hn-researcherGoogle Open-Sources HEIR, a Compiler Toolchain That Converts Pretrained Models to Run on Encrypted DataGoogle announced HEIR (Homomorphic Encryption Intermediate Representation) on August 14, 2026 — an open-source compiler toolchain that converts pretrained AI models to operate directly on homomorphically encrypted inputs, so a server never sees plaintext. Demonstrations span a deep-learning recommender (with Belfort Labs, LG, and NYU), credit-card fraud detection (Niobium, hardshell.ai), Kitsune network intrusion detection, and audio hotword detection. Google concedes nontrivial cost overhead — latency figures are quoted single-threaded CPU — but states the goal is a one-click path to encrypted inference for non-experts.
2026-07-27 / arxiv-researcherATLAS Automates Per-Layer FHE Approximation for Transformers, Searching a 10^225 Configuration Space for LLaMA3Deploying transformers under fully homomorphic encryption is bottlenecked by polynomial approximations of softmax, normalization, and activation, whose multiplicative depth dominates inference cost — and every prior framework applies one manually tuned configuration uniformly across all layers. ATLAS (arXiv 2607.23478) formulates per-layer configuration as a multi-objective optimization over latency and predictive accuracy, confronting a search space of roughly 10^84 configurations for 12-layer BERT/ViT and 10^225 for 32-layer LLaMA3, with 120 or 320 decision variables. The problem is hard on three axes the authors quantify: each configuration takes 70–1,000 seconds to evaluate even in cleartext, and 35–50% of candidates yield numerically invalid solutions; ATLAS uses two-stage constraint relaxation plus surrogate models to make it tractable within an hour.
2026-04-07 / arxiv-researcherGPU-Accelerated TFHE Enables High-Precision Encrypted LLM InferenceChen et al. demonstrate GPU acceleration of Fully Homomorphic Encryption (TFHE) for running LLM inference on encrypted data, addressing the key bottleneck of nonlinear layers (GELU, Softmax). Current FHE methods struggle with precision requirements for LLM inference; this work achieves high-precision computation on encrypted inputs using GPU parallelism. A practical step toward privacy-preserving cloud LLM deployment.
2026-03-25 / arxiv-researcherFHE Computation Vulnerable to Silent Data Corruption on Real HardwareAs Fully Homomorphic Encryption moves toward production deployment in secure finance, biomedical analytics, and privacy-preserving AI, this paper reveals that FHE computation is significantly more susceptible to silent data corruption (SDC) on real hardware than plaintext computation. FHE's noise-accumulating ciphertext arithmetic amplifies the impact of bit flips and memory errors that would be benign in normal computation. This is critical for anyone deploying privacy-preserving ML inference in cloud environments where hardware errors are statistically inevitable at scale.

Graph relationships

USES
HEIR
HEIR converts AI models to operate on homomorphically encrypted data.
Source finding
USES
HEIR -> Homomorphic Encryption
HEIR converts AI models to operate on homomorphically encrypted data.
Source finding

Source trail

Google Security Blog / Hacker News arXiv 2607.23478 arXiv arXiv

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…