SCHEMA reveals that hallucinations concentrate at a small set of highly connected knowledge hubs, and that final-answer accuracy decouples from trajectory honesty; models often reach correct conclusions through structurally flawed reasoning.
Abstract
Large language model (LLM) agents are increasingly deployed in scientific research, where reliability is critical and the underlying knowledge is densely interconnected. In such settings, hallucinations are particularly damaging: a single erroneous claim on a foundational concept can propagate through multi-step reasoning and corrupt entire trajectories. Existing hallucination benchmarks largely operate at the surface level, treating facts in isolation and relying on uniform accuracy metrics that ignore this topological structure. We address this gap with SCHEMA, the first evidence-grounded, topology-aware evaluation framework for hallucinations in scientific agents. SCHEMA automatically constructs scientific concept graphs from benchmark seeds and literature evidence, synthesizes graph-grounded tasks spanning claim verification, multi-hop reasoning, open-ended explanation, and experimental code generation, and evaluates agents with two complementary diagnostics. A trajectory hallucination pipeline audits intermediate reasoning at scale via a topology-weighted severity score, while a multi-agent counterfactual attribution module pinpoints the causal mechanism behind selected failures. SCHEMA reveals that hallucinations concentrate at a small set of highly connected knowledge hubs, and that final-answer accuracy decouples from trajectory honesty; models often reach correct conclusions through structurally flawed reasoning. These results indicate that for high-stakes scientific applications, terminal accuracy alone is an insufficient signal of agent reliability, motivating mechanism-level evaluation grounded in knowledge topology. Code is available at https://github.com/circles-post/SCHEMA.
As large language models are increasingly deployed as tool-augmented legal agents, they introduce agentic hallucinations where tool-call and reasoning errors cascade into fabricated holdings and miscited authority. However, existing legal benchmarks evaluate only single-turn QA with outcome-level metrics, while agentic hallucination benchmarks lack legal-specific diagnostic capability. Neither answers to what extent and how a legal agent hallucinates along its trajectory. To address these limitations, we introduce LexAgentHallu, a legal agentic hallucination benchmark designed to evaluate to what extent and how legal agents fail along multi-step trajectories. Built through a four-stage expert-in-the-loop pipeline, LexAgentHallu contains 3414 instances across 17 legal categories and 6 task types. Each instance is annotated under a dual-layer hallucination taxonomy of 7 high-level categories and 27 fine-grained subclasses, covering both substantive errors and agent-procedural failures. We further design fine-grained metrics that quantify to what extent and localize how each failure occurs along an agent's execution path. Our evaluation across 18 proprietary and open-source agents uncovers a Right-Answer-Wrong-Reason effect and reveals that hallucination subclasses cluster rather than scatter, forming distinct agentic framework, legal task, and category profiles. These findings, invisible to outcome-level evaluation, validate the diagnostic power of LexAgentHallu for evaluating agentic hallucination in law.
Yu-Jin Zhou, Min Zheng, Chuxue Cao et al.· 0 citations
A self-reflective framework in which an LLM generates an answer, identifies claims that may be uncertain, performs an internal verification stage, and revises the response before delivery is proposed.
Priti Sharma, Sachin Sharma· Iconic research and engineer...· 0 citations
HallDetect, a lightweight, reference-free, and black-box framework for hallucination detection, is presented, a lightweight, reference-free, and black-box framework for hallucination detection that is evaluated not only on summarization but across a broader range of source-grounded generation settings.
Achir Oukelmoun, N. Semmar, Gäel de Chalendar· 0 citations
The Latent Critic is introduced, a lightweight low-rank adapter that operates concurrently with a frozen base LLM's generation to actively restructure the transformer's residual stream---amplifying latent grounding signals and translating them into localized, natural language feedback within a single sequence.
AI models, including large language models, are increasingly integrated into scientific discovery workflows, yet they remain prone to hallucination. In experimental sciences, such errors translate directly into failed wet-lab validations and wasted resources; in self-improving agentic systems, confident errors risk being reinforced rather than corrected. Retrosynthesis provides a representative example of this failure mode: existing models can generate chemically plausible routes, but cannot reliably determine which routes are experimentally feasible. We define \textbf{Chemical Hallucination} as a route that appears valid yet fails under competing reactive sites, unresolved selectivity, or missing mechanistic support, a failure largely invisible to the Recall@$K$ metric. We introduce \textbf{SureRoute}, a chemical verifier-anchored retrosynthesis platform that suppresses Chemical Hallucination. SureRoute combines a multi-model ensemble, data asset retrieval, and \textbf{ChemHarness}, an executable chemical intuition engine for route verification and reliability-first ranking. On a benchmark of 350 real-world industrial targets, SureRoute reaches 74.3\% recall@1, 2.2--3.5$\times$ that of seven single-step models and three frontier LLMs, while cutting top-1 Chemical Hallucination to 4.6\%, a 4--6$\times$ reduction relative to frontier LLMs. As a model-agnostic reranker, ChemHarness drives detectable hallucination toward near-zero across arbitrary backbone candidates. SureRoute shows that reliable scientific AI requires not only strong generation, but executable verification.
Jieli Zhou, Nai-Wu Chen, Long-Zhang Liu et al.· 0 citations
A concise two-axis framework that integrates an “intrinsic-extrinsic” distinction in source attribution introduced by Ji et al. with a “faithfulness-factuality” distinction in contextual grounding surveyed is presented, yielding four clearly defined hallucination types applicable across tasks, modalities and architectures.
Misbah Khan, Preston Billion-Polak, T. Khoshgoftaar· IEEE Access· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
