Skip to main content

VeryNews — A Multi-Agent System for Automated News Authenticity Verification

Authors: Thomas, Peter
Original source: Research paper on automated fact-checking systems

Overview​

Misinformation and disinformation in online news pose significant risks to public trust and democratic discourse. VeryNews addresses this challenge through an innovative multi-agent system that automates news authenticity verification. The system uses nine specialized agents that collaborate to deliver transparent, evidence-grounded verdicts in roughly 18.7 seconds per article.

Key Highlights​

  • Multi-agent architecture: Nine specialized agents work together, each optimized for specific tasks like translation, evidence retrieval, and expert analysis.
  • Evidence-grounded reasoning: All judgments are anchored in verifiable evidence from trusted sources (NYTimes, Reuters, AP, BBC), avoiding reliance on parametric knowledge.
  • Explainable AI: Comprehensive transparency through structured 5W1H extraction, evidence provenance tracking, and human-readable reasoning chains.
  • Cross-lingual capabilities: Seamlessly processes multilingual content through intelligent translation and cultural context preservation.
  • High accuracy: Achieves 87% macro-F1 score, significantly outperforming single-agent LLM approaches (74%) and traditional methods (65%).

Visual Insight​

VeryNews agent architecture breakdown

The architecture diagram illustrates how nine specialized agents collaborate in the fact-checking pipeline, from initial translation through evidence aggregation to final judgment.

System Workflow​

  1. Translation Agent translates news content to English if needed.
  2. 5W1H Extraction Agent extracts Who, What, When, Where, Why, and How elements.
  3. Fact-Checking Agent generates targeted search queries and retrieves evidence from trusted sources.
  4. Evidence Aggregation Agent consolidates multi-source evidence into structured chains.
  5. Expert Analysis Agent provides domain-specific analysis and assesses credibility.
  6. Timeliness Agent analyzes timelines based strictly on explicit evidence.
  7. Judgment Agent synthesizes all findings to output a JSON verdict (True/False/Partially True).
  8. Visualization Agent generates timelines and key point summaries.
  9. Report Expert Agent integrates all outputs into a professional Markdown report.

Why It Matters​

VeryNews addresses a critical need in the era of AI-generated content and sophisticated disinformation campaigns. By combining the scalability of automated systems with the rigor of evidence-based verification, it provides a practical tool for journalists, fact-checkers, and platforms working to maintain information integrity. The system's transparency ensures users can understand not just the verdict, but the reasoning and evidence behind it.

Try VeryNews App https://verynews.org/

Back to Research Portfolio