# Study shows AI leaderboard rankings can be gamed

> Researchers find that preference‑based model rankings can be boosted by up to 45 spots using simple manipulation tricks.

Oossa · 2026-11-13 · https://oossa.com/en/study-shows-ai-leaderboard-rankings-can-be-gamed

A paper posted to arXiv on 10 Nov 2026 reveals that arena‑style leaderboards for language models are easy to game. The authors demonstrate three attacks – strategic voting, prompting, and nomination – that can each raise a model’s rank by about 15‑20 places. When combined, the tricks can move a model up roughly 45 spots on the leaderboard. The paper also outlines safeguards, but notes the problem applies to any system using the Bradley‑Terry model for pairwise preferences.

## The facts

- Boosts of up to 15‑20 places per attack, 45 places combined
- Paper submitted 10 Nov 2026

## Why it matters

If unchecked, these hacks could mislead developers and users about which language model truly performs best.

## Sources & references

1. [Leaderboard Hacking: Preference-Based Model Evaluations are Vulnerable to Manipulation](https://arxiv.org/abs/2611.14769) – arXiv, 2026-11-13

Last updated: 2026-11-13