PulseAugur
EN
LIVE 13:08:02

New datasets and surveys tackle multimodal AI adversarial attacks · 2 sources tracked

By PulseAugur Editorial · [4 sources] ·

Two new research papers address the growing challenge of adversarial attacks on AI models, particularly vision-language models (VLMs). The first paper, "Adversarial Diffusion Across Modalities," surveys existing attacks and defenses, proposing a unified framework and identifying weaknesses in current research. The second paper, "PHANTOM," introduces a large-scale, open-source dataset of pre-generated adversarial attacks for VLMs, aiming to lower the barrier for researchers studying model robustness and safety. Both efforts highlight the need for more reproducible and comprehensive evaluations of AI systems against malicious inputs. AI

IMPACT These resources aim to improve the robustness and safety evaluations of vision-language models by providing unified frameworks and accessible datasets for adversarial research.

RANK_REASON Two research papers published on arXiv introducing a survey of adversarial attacks and a new dataset for evaluating vision-language models.

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 4 sources. How we write summaries →

New datasets and surveys tackle multimodal AI adversarial attacks · 2 sources tracked

COVERAGE [4]

  1. arXiv cs.CL TIER_1 English(EN) · Abrar Alotaibi, Moataz Ahmed ·

    Adversarial Diffusion Across Modalities: A Fusion Survey of Attacks, Defenses, and Evaluation for Text, Vision, and Vision-Language Models

    arXiv:2606.26566v1 Announce Type: cross Abstract: Adversarial evaluation of AI systems has matured along four largely disconnected tracks: diffusion-based attacks on text and large language models (LLMs), diffusion-based attacks on image classifiers, jailbreak pipelines against v…

  2. arXiv cs.CL TIER_1 English(EN) · Moataz Ahmed ·

    Adversarial Diffusion Across Modalities: A Fusion Survey of Attacks, Defenses, and Evaluation for Text, Vision, and Vision-Language Models

    Adversarial evaluation of AI systems has matured along four largely disconnected tracks: diffusion-based attacks on text and large language models (LLMs), diffusion-based attacks on image classifiers, jailbreak pipelines against vision-language models, and diffusion-based input p…

  3. arXiv cs.AI TIER_1 English(EN) · Simone Gallivanone, Hossein Khodadadi, Mauro Dore, Mauro Medda, Nicola Franco ·

    PHANTOM: A Large-Scale Dataset of Multimodal Adversarial Attacks for Vision-Language Models

    arXiv:2606.24388v1 Announce Type: new Abstract: We introduce a large-scale, open-source dataset of pre-generated adversarial attacks for vision-language models (VLMs). The dataset is designed to be diverse, representative, and practical, extending existing benchmarks by covering …

  4. arXiv cs.AI TIER_1 English(EN) · Nicola Franco ·

    PHANTOM: A Large-Scale Dataset of Multimodal Adversarial Attacks for Vision-Language Models

    We introduce a large-scale, open-source dataset of pre-generated adversarial attacks for vision-language models (VLMs). The dataset is designed to be diverse, representative, and practical, extending existing benchmarks by covering 10 high-level categories and 55 subcategories of…

RELATED ENTITIES

RELATED TOPICS