Researchers have developed a new type of attack on fact-checking systems that use artificial intelligence to verify the accuracy of claims. This 'adversarial persuasion' attack uses language models to rephrase false claims in a way that makes them more convincing and harder for AI fact-checkers to detect. The study found that this type of attack can significantly degrade the performance of fact-checking systems, highlighting the need for more robust defenses against disinformation. --- Why it matters: This matters because it shows how easily fact-checking systems can be manipulated by adversaries using sophisticated language techniques. Engineers and researchers in AI will need to develop new methods to counter these types of attacks and improve the resilience of fact-checking systems. Source: https://arxiv.org/abs/2601.16890