TRUST Agents: A Collaborative Multi-Agent Framework for Fake News Detection, Explainable Verification, and Logic-Aware Claim Reasoning

TRUST Agents is a multi-agent AI framework designed to improve fake news detection and fact verification by combining claim extraction, evidence retrieval, verification, and explainable reasoning. Unlike binary classification approaches, the system generates transparent, human-inspectable reports with logic-aware reasoning for complex claims, though it shows that retrieval quality and uncertainty calibration remain significant challenges in automated fact verification.

TRUST Agents addresses a critical gap in current fact-verification systems: the need for explainability and transparency alongside accuracy. Traditional fake news detection treats verification as binary classification, obscuring the reasoning process and making it difficult for users to understand why a claim was flagged. This framework reconstructs the human verification workflow by decomposing the task into specialized agents that handle claim extraction, evidence retrieval, comparison, and explanation generation separately.

The system's architecture reflects broader advances in multi-agent AI systems where specialized components collaborate on complex problems. By incorporating logic-aware reasoning through conjunction, disjunction, and implication operations, TRUST Agents handles compound claims more effectively than single-pass classifiers. The Delphi-inspired jury component demonstrates how multiple AI perspectives with different specializations can converge on more robust verdicts.

For the misinformation ecosystem, this work has implications for platform trust and user literacy. As AI-generated and manipulated content becomes more sophisticated, automated systems alone cannot maintain credibility without transparent reasoning. Media platforms and fact-checking organizations could leverage such frameworks to provide users with detailed verification traces rather than simple verdicts. The research highlights that performance bottlenecks now center on evidence retrieval quality and confidence calibration rather than architectural complexity.

The comparative evaluation against fine-tuned BERT and RoBERTa models reveals a classic trade-off: supervised encoders achieve stronger raw accuracy, but TRUST Agents offers superior interpretability. As regulatory pressure around AI transparency increases, the ability to explain decisions may become commercially valuable. Future development should focus on improving the retrieval and calibration components identified as key bottlenecks.

• →TRUST Agents prioritizes explainability in fact verification, enabling users to inspect reasoning and evidence citations rather than receiving opaque verdicts.
• →The framework uses specialized agents for claim extraction, retrieval, verification, and explanation generation, mirroring human fact-checking workflows.
• →Logic-aware aggregation through conjunction, disjunction, and implication operators improves handling of complex, compound claims.
• →Evaluation shows supervised fine-tuned models outperform on raw metrics, but TRUST Agents achieves better interpretability and evidence transparency.
• →Evidence retrieval quality and uncertainty calibration emerge as the primary technical bottlenecks limiting trustworthy automated fact verification.