# Textual Response-Aware Guidance for Zero-Shot Adversarial Robustness of Vision-Language Models

Source: https://arxiv.org/abs/2610.40309
Published: 2026-10-29
Status: unreviewed source report

## Source excerpt (up to 25 words)

> Vision-language models (VLMs) exhibit strong zero-shot generalization but remain vulnerable to adversarial attacks. Existing adversarial fine-tuning methods often use pre-trained model guidance to preserve image--text

## Review question

Inspect benchmark version, contamination controls, sample size, held-out tasks, adversarial failures and scoring rules. Higher scores alone do not establish lower production risk.

No independent proof check or claim of peer review.