LLMs as Legal Decision Tools: Persuadability Study
A new arXiv preprint (2604.26233) examines how Large Language Models (LLMs) respond to legal arguments, focusing on their persuadability. The study tests frontier open- and closed-weights LLMs as proposed decision assistants or first-instance decision-makers in judicial and administrative contexts. It explores whether LLMs can engage with and be persuaded by arguments from contending parties without being unduly influenced by advocate skill over case merits. Original experimental results report how argument quality affects LLM legal decisions.
Key facts
- arXiv preprint 2604.26233
- LLMs proposed as legal decision assistants and first-instance decision-makers
- Study examines persuadability of LLMs by legal arguments
- Tests frontier open- and closed-weights LLMs
- Focus on judicial and administrative contexts
- Explores influence of advocate quality on LLM decisions
- Reports original experimental results
- Aims to ensure LLMs decide based on case merits, not advocate skill
Entities
Institutions
- arXiv