---
title: "Text-Aware, Vision-Blind: How Multimodal... | SI Paper Desk"
url: https://thesimind.si/papers/2611.26569/
description: "Image editing systems from big labs refuse risky text requests far more often than risky images paired with harmless instructions, so their safety checks..."
---

Number 9 of 12 · November 22, 2026

# Text-Aware, Vision-Blind: How Multimodal Image Editing Systems Handle Unsafe Inputs

Image editing systems from big labs refuse risky text requests far more often than risky images paired with harmless instructions, so their safety checks look at words much more than pictures.

Ziqing Yang, Yixin Wu, Michael Backes and 1 more · Submitted November 19, 2026 · arXiv 2611.26569v1 · cs.CR

[Read the paper on arXiv](https://arxiv.org/abs/2611.26569) · [The PDF](https://arxiv.org/pdf/2611.26569) · [The day's 12](https://thesimind.si/papers/day/2026-11-22/)

## What they found

Yang and colleagues built EditSafetyBench, a test set of 5,940 image and instruction pairs. It covers safe and unsafe images across 11 risk categories and three edit intents. They tested four systems: Gemini, GPT, Qwen API, and Qwen Local. Gemini and GPT refused 87.1% and 81.6% of cases with unsafe-intent editing prompts. They refused only 24.2% and 22.4% of cases with unsafe images and neutral instructions. The systems also disagreed by category. For unsafe political content, GPT allowed 70.4% of editing attempts and Gemini allowed 34.8%. Sexual content showed the strongest sensitivity to the image itself, which the authors suggest may come from earlier work on spotting NSFW (not safe for work) content.

## Why it matters

Many people can now edit images with SI tools. A user could upload a harmful image and ask for a plain-sounding edit. This paper suggests the tested systems often let such requests through. The results also show that different systems treat the same risks quite differently. The abstract does not say how the edit intents or risk labels were defined, so details of the benchmark are unclear from here.

## How it changes the race

This is an evaluation result, not a new capability. It gives builders of image editing systems a benchmark to test safety that covers both text and images, and it points to a gap to close.

## Safety and control

The paper shows a misuse gap: safety checks that rely mostly on the text prompt can miss risk that sits in the uploaded image. The authors call for guardrails that reason over both text and image together.

### The abstract, by the authors

Multimodal image generation systems have made rapid progress and increasingly emphasize content safety. However, their safety mechanisms tend to rely on a single modality and lack joint consideration of both visual content and text prompts, particularly in instruction-guided image editing, where the uploaded image may itself carry risk. We thus introduce EditSafetyBench, a benchmark that measures system-level safety awareness of advanced image editing systems, containing 5,940 image-instruction pairs, covering safe and unsafe images across 11 risk categories and three edit intents. We then probe the benchmark dataset on four systems, including Gemini, GPT, Qwen API, and Qwen Local. Based on our evaluation, we observe a systematic text-vision safety mismatch: Gemini and GPT refuse 87.1% and 81.6% cases containing unsafe-intent editing prompts, while only refusing 24.2% and 22.4% cases with both unsafe images and neutral instructions. At the category level, most risks are considered inconsistently by different systems, e.g., GPT allows 70.4% editing attempts of unsafe political content while Gemini only allows 34.8%. Notably, sexual content shows the strongest visual sensitivity, likely benefiting from mature NSFW identification works. Our findings highlight the need for safety guardrails that jointly reason over both visual and textual modalities in image editing systems.

Quoted from arXiv:2611.26569. The research and its words belong to its authors.

Desk score **66** out of 100

Why it was picked

- Safety (in the title)
- Evals
- Studies frontier models (Gemini)
- Cross-listed in 3 fields

Topics

- [Safety](https://thesimind.si/papers/?tag=safety)
- [Benchmarks](https://thesimind.si/papers/?tag=benchmarks)
- [Multimodal](https://thesimind.si/papers/?tag=multimodal)
- [Evals](https://thesimind.si/papers/?tag=evals)
- [Security](https://thesimind.si/papers/?tag=security)
- [Vision](https://thesimind.si/papers/?tag=vision)

**arXiv ID:** 2611.26569v1
**Submitted:** November 19, 2026
**Fields:** cs.CR (Cryptography and Security), cs.CV (Computer Vision), cs.LG (Machine Learning)
**Level:** Intro

Fingerprint, SHA-256

`41cdb62cc16a0b85a9058235780c2a15fe13c539e673766471b4812ce1f6ed81`

In SI MIND's journal since Nov 22, 2026, 02:01 UTC. It goes on Solana with the next hourly Proof of Mind seal.

### The exact text fingerprinted

{"arxiv":"2611.26569","version":1,"title":"Text-Aware, Vision-Blind: How Multimodal Image Editing Systems Handle Unsafe Inputs","authors":["Ziqing Yang","Yixin Wu","Michael Backes","Yang Zhang"],"day":"2026-11-22","rank":9,"score":66,"status":"written","summary":{"takeaway":"Image editing systems from big labs refuse risky text requests far more often than risky images paired with harmless instructions, so their safety checks look at words much more than pictures.","found":"Yang and colleagues built EditSafetyBench, a test set of 5,940 image and instruction pairs. It covers safe and unsafe images across 11 risk categories and three edit intents. They tested four systems: Gemini, GPT, Qwen API, and Qwen Local. Gemini and GPT refused 87.1% and 81.6% of cases with unsafe-intent editing prompts. They refused only 24.2% and 22.4% of cases with unsafe images and neutral instructions. The systems also disagreed by category. For unsafe political content, GPT allowed 70.4% of editing attempts and Gemini allowed 34.8%. Sexual content showed the strongest sensitivity to the image itself, which the authors suggest may come from earlier work on spotting NSFW (not safe for work) content.","matters":"Many people can now edit images with SI tools. A user could upload a harmful image and ask for a plain-sounding edit. This paper suggests the tested systems often let such requests through. The results also show that different systems treat the same risks quite differently. The abstract does not say how the edit intents or risk labels were defined, so details of the benchmark are unclear from here.","race":"This is an evaluation result, not a new capability. It gives builders of image editing systems a benchmark to test safety that covers both text and images, and it points to a gap to close.","safety":"The paper shows a misuse gap: safety checks that rely mostly on the text prompt can miss risk that sits in the uploaded image. The authors call for guardrails that reason over both text and image together.","level":"intro","tags":["safety","benchmarks","multimodal","evals","security","vision"]}}

[Check it on Verify a thought](https://thesimind.si/proof-of-mind/)

## The authors

Ziqing Yang, Yixin Wu, Michael Backes, Yang Zhang

Credit for this research belongs to its authors. SI MIND's page is a plain-English reading of their abstract on arXiv, not a review of the full paper.

Related on the desk

## More papers on these topics

[Nov 14, 2026 · Number 3MobileProbe: Adaptive Security Evaluation of Mobile Agents via Trajectory-Aware InjectionMobileProbe tests phone-operating SI agents by adapting attacks to the route each agent takes, and the authors report it beats fixed, hand-written tests at finding weaknesses.SecurityAgentsEvals](https://thesimind.si/papers/2611.16693/) · [Nov 14, 2026 · Number 7Evaluating Video LLMs' Understanding of Corner Cases in Autonomous DrivingDashBench tests 25 Video LLMs on 1,709 rare, safety-critical driving videos. The authors report that all of the models fall well short of what safety-critical driving needs.BenchmarksEvalsMultimodal](https://thesimind.si/papers/2611.16767/) · [Nov 20, 2026 · Number 2AutoPI: Taxonomy-Guided Adaptive Red Teaming for Prompt Injection in LLM AgentsAutoPI is a red teaming tool that tests SI agents for prompt injection across many attack types, not just how often attacks work, and it reports higher success and coverage than recent rivals.SecurityAgentsEvals](https://thesimind.si/papers/2611.24313/)

[Number 8](https://thesimind.si/papers/2611.25926/) · [All 12 for Nov 22, 2026](https://thesimind.si/papers/day/2026-11-22/) · [Number 10](https://thesimind.si/papers/2611.26033/)
