mlsanityData to 14 Oct 2026

GRAID: Synthetic Data Generation with Geometric Constraints and Multi-Agentic Reflection for Harmful Content Detection

arXiv 2508.17057 ↗First posted 23 Aug 2025cs.CL cs.CR cs.LGcited by 1

On the map: LLM agents and tool use → Agent threats, guardrails and governance

also in 2 more
Abstract (arXiv)

We address the problem of data scarcity in harmful text classification for guardrailing applications and introduce GRAID (Geometric and Reflective AI-Driven Data Augmentation), a novel pipeline that leverages Large Language Models (LLMs) for dataset augmentation. GRAID consists of two stages: (i) generation of geometrically controlled examples using a constrained LLM, and (ii) augmentation through a multi-agentic reflective process that promotes stylistic diversity and uncovers edge cases. This combination enables both reliable coverage of the input space and nuanced exploration of harmful content. Using two benchmark data sets, we demonstrate that augmenting a harmful text classification dataset with GRAID leads to significant improvements in downstream guardrail model performance.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use. Builds on 23 of its 42 references that are in the map. First posted on arXiv 23 Aug 2025.

Builds on

23 of 42 references.

References · 23

  1. 2017Attention Is All You Needcited by 60,361
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  3. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
  4. 2018UMAP: Uniform Manifold Approximation and Projection for Dimension Reductioncited by 2,898
  5. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
  6. 2015A Diversity-Promoting Objective Function for Neural Conversation Modelscited by 1,077
  7. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 860
  8. 2023Tree of Attacks: Jailbreaking Black-Box LLMs Automaticallycited by 672
  9. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 560
  10. 2024Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inferencecited by 554
  11. 2024WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMscited by 496
  12. 2021A Survey of Data Augmentation Approaches for NLPcited by 409
11 more
  1. 2024On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Surveycited by 244
  2. 2018A System for Massively Parallel Hyperparameter Tuningcited by 202
  3. 2024ALERT: A Comprehensive Benchmark for Assessing Large Language Models' Safety through Red Teamingcited by 98
  4. 2022AugESC: Dialogue Augmentation with Large Language Models for Emotional Support Conversationcited by 79
  5. 2023AART: AI-Assisted Red-Teaming with Diverse Data Generation for New LLM-powered Applicationscited by 54
  6. 2024Beyond Answers: Transferring Reasoning Capabilities to Smaller LLMs Using Multi-Teacher Knowledge Distillationcited by 47
  7. 2023The Parrot Dilemma: Human-Labeled vs. LLM-augmented Data in Classification Taskscited by 35
  8. 2023Selective In-Context Data Augmentation for Intent Detection using Pointwise V-Informationcited by 24
  9. 2025Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignmentcited by 14
  10. 2024Knowledge Distillation Using Frontier Open-source LLMs: Generalizability and the Role of Synthetic Datacited by 12
  11. 2022Constructing Highly Inductive Contexts for Dialogue Safety through Controllable Reverse Generationcited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026AdaGuard: Enhancing Safety and Policy Compliance with Reasoning-Enabled LLM-As-A-Judge Guardrailscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024When in Doubt, Cascade: Towards Building Efficient and Capable Guardrails
  2. 2025Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails
  3. 2026Opir: Efficient Multi-Task Safety Classification for Toxicity, Jailbreaks, Hate Speech, and Harmful Content
  4. 2026CourtGuard: A Model-Agnostic Framework for Zero-Shot Policy Adaptation in LLM Safety
  5. 2024HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models
  6. 2026Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models
  7. 2025HarmTransform: Transforming Explicit Harmful Queries into Stealthy via Multi-Agent Debate
  8. 2025OpenGuardrails: A Configurable, Unified, and Scalable Guardrails Platform for Large Language Models