# Self‑grounded verification reduces agreement bias in multimodal LLMs

> Researchers introduce SGV, a two‑step method that helps multimodal language models spot wrong actions more reliably, improving downstream tasks by up to 25 percentage points.

Oossa · 2026-11-13 · https://oossa.com/en/self-grounded-verification-reduces-agreement-bias-in-multimodal-llms

Moises Andrade and colleagues present SGV, a verification technique for multimodal LLMs that first asks the model to list what correct behavior looks like, then judges a specific action against that list. The paper shows the method cuts the models’ tendency to over‑approve actions – called agreement bias – and raises failure‑detection rates by up to 25 percentage points and overall accuracy by 14 percentage points. Applied to a GUI specialist in OSWorld, a diffusion policy in RoboMimic, and a ReAct agent in VisualWebArena, the verifiers improve task completion by about 20 percentage points over the previous best results.

## The facts

- SGV improves failure detection by up to 25 percentage points
- Task completion gains of ~20 percentage points on OSWorld, RoboMimic and VisualWebArena

## Why it matters

Better verification lets AI agents avoid false confidence, leading to more reliable performance for users of visual‑web and robotics tools.

## Sources & references

1. [Let's Think in Two Steps: Mitigating Agreement Bias in MLLMs with Self-Grounded Verification](https://arxiv.org/abs/2507.11662) – arXiv cs.MA, 2026-11-13

Last updated: 2026-11-13