According to the abstract for arXiv:2610.24873, the proposed one-relation attack (ORA) is a cognitive attack that targets a single implicit warrant—the reasoning link that connects evidence to a conclusion. The central question is whether one sentence can manipulate an agent's belief by exploiting that narrow inferential connection.
The title states the core finding: one truthful relation is enough to mislead LLM agents. This matters because the attack does not rely on falsehoods or long adversarial prompts; it uses a single truthful statement placed at the right point in the reasoning chain.
Since the source is only the abstract, the article does not include experimental details or comparisons with other attack methods. The key claim is that a minimal, truthful relation can be sufficient to shift an agent's belief.