mlsanityData to 14 Oct 2026

Can We Trust a Black-box LLM? LLM Untrustworthy Boundary Detection via Bias-Diffusion and Multi-Agent Reinforcement Learning

arXiv 2604.05483 ↗First posted 7 Apr 2026cs.AI cs.CLcited by 0

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks

Abstract (arXiv)

Large Language Models (LLMs) have shown a high capability in answering questions on a diverse range of topics. However, these models sometimes produce biased, ideologized or incorrect responses, limiting their applications if there is no clear understanding of which topics their answers can be trusted. In this research, we introduce a novel algorithm, named as GMRL-BD, designed to identify the untrustworthy boundaries (in terms of topics) of a given LLM, with black-box access to the LLM and under specific query constraints. Based on a general Knowledge Graph (KG) derived from Wikipedia, our algorithm incorporates with multiple reinforcement learning agents to efficiently identify topics (some nodes in KG) where the LLM is likely to generate biased answers. Our experiments demonstrated the efficiency of our algorithm, which can detect the untrustworthy boundary with just limited queries to the LLM. Additionally, we have released a new dataset containing popular LLMs including Llama2, Vicuna, Falcon, Qwen2, Gemma2 and Yi-1.5, along with labels indicating the topics on which each LLM is likely to be biased.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 15 of its 38 references that are in the map. First posted on arXiv 7 Apr 2026.

Builds on

15 of 38 references.

References · 15

  1. 2017Graph Attention Networkscited by 8,923
  2. 2017Inductive Representation Learning on Large Graphscited by 6,320
  3. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,148
  4. 2013Playing Atari with Deep Reinforcement Learningcited by 3,274
  5. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,255
  6. 2020Language (Technology) is Power: A Critical Survey of "Bias" in NLPcited by 956
  7. 2016Semantics derived automatically from language corpora contain human-like biasescited by 901
  8. 2017Rainbow: Combining Improvements in Deep Reinforcement Learningcited by 786
  9. 2020CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Modelscited by 710
  10. 2019The Woman Worked as a Babysitter: On Biases in Language Generationcited by 541
  11. 2019On Measuring Social Biases in Sentence Encoderscited by 398
  12. 2019Social Bias Frames: Reasoning about Social and Power Implications of Languagecited by 380
3 more
  1. 2020Measuring and Reducing Gendered Correlations in Pre-trained Modelscited by 224
  2. 2017Fairness in Machine Learning: Lessons from Political Philosophycited by 149
  3. 2016Deep-Learned Collision Avoidance Policy for Distributed Multi-Agent Navigationcited by 15

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Deep Learning-based Method for Expressing Knowledge Boundary of Black-Box LLM
  2. 2026Knowledge Boundary Discovery for Large Language Models
  3. 2025Teaching LLMs to Abstain via Fine-Grained Semantic Confidence Reward
  4. 2025Learning the Boundary of Solvability: Aligning LLMs to Detect Unsolvable Problems
  5. 2025Bias Testing and Mitigation in Black Box LLMs using Metamorphic Relations
  6. 2025General-Reasoner: Advancing LLM Reasoning Across All Domains
  7. 2026Can LLM Reasoning Be Trusted? A Comparative Study: Using Human Benchmarking on Statistical Tasks
  8. 2025Single LLM Debate, MoLaCE: Mixture of Latent Concept Experts Against Confirmation Bias