Large language models are increasingly used to support decisions across many domains. A new arXiv paper asks whether these models are biased in favour of artificial intelligence itself. The authors investigate whether LLMs systematically prefer AI when acting as decision-support tools.

The study, posted as arXiv:2601.13749, examines the question through three experiments. The abstract does not describe the design or results of those experiments, so the findings are not yet available from the summary alone.

The question matters because a pro-AI bias could shape the recommendations these models give to human users. The paper highlights a potential blind spot in how LLMs are evaluated and deployed.