Research map / Large language models / Agents, code and math
FHE-Coder: Secure Agentic Code Generation for Fully Homomorphic Encryption
Mayank Kumar, Jiaqi Xue, Mengxin Zheng, Qian Lou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| run full FHE-Coder pipeline instead of baseline agent | GPT-5 TFHE agent · TFHE MatMul/Softmax/Attention/Transformer | pass@1 (func_sec): 0.00 -> 1.00 (MatMul) | Improved |
| add Prompt Formalizer to baseline (ablation) | GPT-5 baseline · TFHE Vector Addition | pass@1 (func_sec): 0.0 -> 0.6 | Improved |
| add API RAG retriever on top of Formalizer (ablation) | BAS + FP · TFHE Vector Addition | pass@1 (func_sec): 0.6 -> 0.8 | Improved |
| add Security Verifier feedback loop on top of FP and RAG (ablation) | BAS + FP + RAG · TFHE Vector Addition | pass@1 (func_sec): 0.8 -> 1.0 | Improved |
| add structured decomposition (compose verified primitives) (ablation) | FHE-Coder · TFHE Matrix-Vector, CNN | pass@1 (func_sec): near-zero -> 0.70 (Mat-Vec), 0.60 (CNN) | Improved |
| use expert-enriched docstring RAG instead of vanilla RAG (ablation) | FHE-Coder RAG retriever · TFHE ReLU, MatMul | pass@1 (sec): ReLU sec 0.20 with vanilla RAG; MatMul sec 0.00 | Improved |
| swap TFHE parameter solver and docs for CKKS versions | FHE-Coder · CKKS Softmax/Attention/Transformer | pass@1 (func_sec): 0.00 -> 0.60 (Softmax), 1.00 (Attention) | Improved |
| FHE-Coder on CKKS MatMul vs baseline functional correctness | GPT-5 CKKS · CKKS MatMul | pass@1 (func): 1.00 -> 0.80 | Hurt |
| FHE-Coder on Transformer workloads | GPT-5 · TFHE Transformer | pass@1 (func_sec): 0.00 -> 0.80 per Table 4; about 0.4 per text | Mixed |
| use stronger base LLM (GPT-5, Gemini-2.5-Pro vs DeepSeek-V3.1, Qwen3-Coder) | FHE-Coder · TFHE tasks across four LLMs | pass@1 (func_sec) | Improved |
Closest papers by meaning
- EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (ICML 2025 · similarity 0.88)
- Encryption-Friendly LLM Architecture (ICLR 2025 · similarity 0.86)
- SecureClaw: Clawing Back Control of LLM Agents (NeurIPS 2026 · similarity 0.85)
- Verify0: Can AI Agents Build Formally Verified Software Repositories? (NeurIPS 2026 · similarity 0.84)
- SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks (NeurIPS 2025 · similarity 0.84)
- DOMAINEVAL: An Auto-Constructed Benchmark for Multi-Domain Code Generation (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
