New S-GAP framework enhances stealthy adversarial attacks on GraphLLMs

Researchers have introduced S-GAP, a novel framework designed to improve adversarial attacks against Graph Large Language Models (GraphLLMs). S-GAP unifies textual and structural perturbations into a single 'Graph Attack Primitive' (GAP) to enhance attack stealth and effectiveness. The system uses a multi-agent approach for iterative refinement and surrogate evaluation, demonstrating an average 7.02% improvement in attack success rate over existing methods across various datasets and models. AI

IMPACT Introduces a more effective method for evaluating and potentially improving the robustness of GraphLLMs against adversarial manipulation.

RANK_REASON Academic paper detailing a new method for adversarial attacks on GraphLLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Graph Attack Primitive
• GraphLLMs
• S-GAP

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →