UQLM: Uncertainty Quantification for Language Models, is a Python package for UQ-based LLM hallucination detection

[NeurIPS 2025] SECA: Semantically Equivalent and Coherent Attacks for Eliciting LLM Hallucinations

KIPRIS 특허·실용신안·상표·디자인 검색 MCP — 자유검색/항목검색/출원인/권리자/서지상세 7개 도구 | KIPRIS Korean patent·utility·trademark·design search → 7 MCP tools

[ICML 2026] REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations

(개요) 국가법령정보센터와 알리오의 공공기관 내부규정을 검색·비교·분석하는 MCP. (도구) 법제처 87 + ALIO 공공기관 규정 23 = 110개 MCP 도구. (데이터) 1,600 법률, 10,000 행정규칙, 수만건 판례, 344개 공공기관 35,000 내부규정.

RAG Hallucination Detecting By LRP.

CRoPS (TMLR)

Novel Hallucination detection method

A formally-grounded governance framework for Kilo Code. Establishes explicit policy foundations and experimental guardrails across Kilo’s Architect and Code modes to eliminate model hallucination, unauthorized scope creep, and unverified architecture drift.

Make your AI coding agent prove every claim with a file:line citation, then machine-check each one offline. A verified-citation gate (CLI + MCP server + GitHub Action) that catches LLM hallucinations before merge.

Research paper on how agentic debate pipelines can be constructed to reduce hallucinations in LLMs with open-source and commercial models

Semi-supervised pipeline to detect LLM hallucinations. Uses Mistral-7B for zero-shot pseudo-labeling and DeBERTa for efficient classification.

Build your own open-source REST API endpoint to detect hallucination in LLM generated responses.

A theoretical framework for embedding lightweight, controllable AI into enterprise information systems (ERP, finance, supply chain) without relying on general-purpose LLMs. L0-L5 evolution model + hallucination control + human-in-the-loop.（2026-06-04 | PSSXiv:202606.02680V1）

법고개 — LLM·문서 속 한국 판례/법령 인용의 진위를 실시간 검증하는 Chrome 확장 · Chrome extension that flags AI-fabricated Korean case-law citations in real time

ESA Impostor Hunt (Secure Your AI) — detects hallucinated and data-poisoned LLM outputs from faithful ones. Public score: 0.90663 using CatBoost + SciBERT embeddings, trained on 95 samples against 1,068 test pairs. Features GPT-2 perplexity, bigram Jaccard, KL divergence, burstiness; tuned via Optuna + 5-fold CV.

Official PyTorch implementation of a mechanistic interpretability framework for Self-Explaining LLMs. Generates real-time mathematical provenance and logical explanations for model outputs to detect, track, and eliminate hallucinations in critical domains.

Verify, repair, and audit LLM-hallucinated BibTeX citations against authoritative scholarly sources.

Source code for the paper: A Hallucination Mitigation Scheme in Security Policy Generation with Large Language Models (Accepted at KICS Winter Conference 2026)