# New benchmark: AI guardrails miss most GEO misinformation attacks

Published: 2026-09-03T01:48:53.125Z · Source: arXiv GEO (https://arxiv.org/abs/2609.02316v1)
Source date: 2026-09-02

A new arXiv preprint introduces Counter-GEO-Bench, a benchmark for testing whether AI safety guardrails catch misinformation delivered through generative engine optimization (GEO) — ordinary-looking documents crafted so large language models retrieve and repeat distorted claims. Across 247 human-verified queries and three victim LLMs the paper does not name, the authors found three off-the-shelf guardrails (Granite Guardian, Llama Guard 3, NeMo Self-Check Fact-Checking) cut attack success by at most 5.7% relative, with Granite Guardian's reduction not statistically significant — because the guardrails are built to catch policy violations, not fluent but factually wrong text. The authors' own proposed defense, C-GEO Guard, cut attack success by 47.6% relative with near-zero utility loss. The paper is accepted to EMNLP 2026.

Why it matters: It's the first controlled evidence that GEO techniques built to win citations can also be used to push distorted answers past today's safety guardrails, which are built to catch policy violations rather than fluent, factually wrong text.

## What this answers

**Can AI safety guardrails stop misinformation delivered through GEO-optimized content?**

Not well, per a new arXiv benchmark: three off-the-shelf guardrails (Granite Guardian, Llama Guard 3, NeMo Self-Check Fact-Checking) cut attack success by at most 5.7% relative, and Granite Guardian's reduction wasn't statistically significant.

**What is Counter-GEO-Bench?**

A new benchmark pairing 247 human-verified queries with information-preserving and information-distorting GEO rewrites, used to test how well AI defenses catch GEO-based misinformation across three large language models.


Canonical: https://anythingengineoptimization.com/item/2026-09-03-new-benchmark-ai-guardrails-miss-most-geo-misinformation-attacks/
From Anything Engine Optimization (AEO Wire) — https://anythingengineoptimization.com/ · Standards: https://anythingengineoptimization.com/standards/