arXiv:2607.00415v1 Announce Type: new Abstract: Authority bias poses a critical safety concern in language models: models systematically prioritize social cues from authority figures over factual consistency, swaying their answers based on source credibility rather than evidence. We mechanistically investigate this phenomenon using a controlled medical QA setting, where hints suggesting incorrect answers are attributed to personas of varying expertise. Across Llama-3.1-8B, Qwen3-8B, and Gemma-2-9B, we find that models respond in a graded manner proportional to perceived authority, a hierarchy
The increasing deployment of LLMs into critical applications makes understanding their biases, like authority sycophancy, a pressing concern for safety and reliability.
This research highlights a fundamental flaw in current LLM architectures, where perceived authority can override factual accuracy, posing significant risks for trust and decision-making.
Our understanding of LLM reliability shifts from purely factual recall to recognizing the susceptibility of models to social cues, demanding new alignment and fine-tuning strategies.
- · AI safety researchers
- · Developers of robust LLM evaluation frameworks
- · Ethical AI consultants
- · LLMs without strong alignment against authority bias
- · Users relying on LLMs for fact-checking without critical oversight
- · Applications where perceived authority can be manipulated
Further research and development will focus on mitigating authority bias in large language models.
New regulatory guidelines and industry standards may emerge to address LLM sycophancy in high-stakes applications.
Public trust in AI systems could erode if these biases lead to significant real-world failures or misinformation campaigns facilitated by AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL