# A Systematic Study of Cross-Modal Typographic Attacks on Audio-Visual Reasoning

Source: [A Systematic Study of Cross-Modal Typographic Attacks on Audio-Visual Reasoning](<https://www.sota2.com/research/paper/69d48f7b7d0172140ec742a9>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-12T06:00:01.362Z

Authors: Tianle Chen, Deepti Ghadiyaram

Published: 2026-04-05T06:32:08.000Z

## Abstract

As audio-visual multi-modal large language models (MLLMs) are increasingly deployed in safety-critical applications, understanding their vulnerabilities is crucial. To this end, we introduce Multi-Modal Typography, a systematic study examining how typographic attacks across multiple modalities adversely influence MLLMs. While prior work focuses narrowly on unimodal attacks, we expose the cross-modal fragility of MLLMs. We analyze the interactions between audio, visual, and text perturbations and reveal that coordinated multi-modal attack creates a significantly more potent threat than single-modality attacks (attack success rate = $83.43\\%$ vs $34.93\\%$).Our findings across multiple frontier MLLMs, tasks, and common-sense reasoning and content moderation benchmarks establishes multi-modal typography as a critical and underexplored attack strategy in multi-modal reasoning. Code and data will be publicly available.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Audio Visual Question Answering on Music-AVQA (Robustness Evaluation)](<https://www.sota2.com/research/sota/audio-visual-question-answering-on-music-avqa-robustness-evaluation.md>): Audio-Visual Question Answering; MUSIC-AVQA; Music-AVQA Clean Accuracy: not recorded.

- [Audio Question on MMA-Bench](<https://www.sota2.com/research/sota/audio-question-on-mma-bench.md>): Audio Question; MMA-Bench; Accuracy Clean: not recorded.

- [Audio-Visual Question on WorldSense (Robustness Evaluation)](<https://www.sota2.com/research/sota/audio-visual-question-on-worldsense-robustness-evaluation.md>): Audio-Visual Question; WorldSense; Accuracy (Clean): not recorded.

- [Visual Question Answering on MMA-Bench (Adversarial Robustness)](<https://www.sota2.com/research/sota/visual-question-answering-on-mma-bench-adversarial-robustness.md>): Visual Question; MMA-Bench; Clean Accuracy: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.03995>)

[PDF](<https://arxiv.org/pdf/2604.03995.pdf>)