News copies trick large language models into trusting false claims
Copies Are Not Confirmations: How News Duplication Fools LLM Judges and Continual Learners
Machine LearningComputation and Language
Summary
Many news articles repeat the same story copied from one source, which can trick AI systems into thinking multiple sources confirm the story. The authors found that large language models often accept false claims more when seeing repeated copies than a single report. Some models ignore notes about the duplication, while others use provenance info to avoid being fooled. They also tested ways to detect reworded copies by comparing named entities and found methods to reduce AI acceptance of false claims from copies. Counting sources alone is not enough to ensure reliability.
What this means in practice
- •For news verification teams: Filter out repeated copies of news reports to reduce AI errors in assessing claim veracity from multiple outlets.
- •For data pipeline engineers: Design continual learning systems that avoid counting duplicate news reports as independent information to improve accuracy.
Authors
Abdelkader Metales, Miriam Grabenheinrich, Aicha Metales
Abstract
Many news reports of a claim are copies of one source: wire stories, content shared within a media group, syndicated features and press releases. We ask whether LLM judges and continual learners treat such copies as independent confirmations. In 200,000 English CC-News articles, a MinHash detector links 18.9% to a near-copy from another outlet (precision 0.97 once boilerplate is removed); it links 92% of edited reprints to their originals but none of the LLM rewrites by phi4, Llama 3.2 or Gemini 2.5 Flash. Shown three identical reprints of a claim about an unknown person instead of one report, Llama 3.2 3B and phi4 14B raised the share of lies they accepted from 0.06 to 0.71 and from 0.21 to 0.90 and ignored a note that the reports came from one source; Gemini 2.5 Flash already believed a single report, and Gemini 2.5 Pro followed provenance notes and tables. In a simulated continual-learning stream, showing the judge one report per detected chain cut the copied lies it accepted by 0.10-0.24 across four judges, at a mean cost of no more than 4.2 percentage points of accepted truths. Against reworded copies, a test of how many of a report's named entities are absent from an earlier look-alike separates LLM rewrites from 100 labelled independent reports of the same event with AUC 0.98-0.99 and wrongly flags one; entity overlap alone does as well, TF-IDF and a current embedding model come close, and the test catches few natural same-source derivatives. With the miss patterns of generated rewrite networks, it cut the reworded-copy lies accepted from 0.67 to 0.20 (0.40 when the rewriter invents details). Counting reports or chains alone is not a reliable defence, the effect of provenance information depends on the judge, and colluding sources remain the open problem. Code and labels: https://github.com/Metalesaek/copies-experiment