RAG-Pull: Turning Retrieval into a Code-Injection Channel via Invisible Unicode Perturbations
📰 ArXiv cs.AI
Learn how RAG-Pull attacks inject malicious code into LLMs via invisible Unicode perturbations, compromising model reliability and trustworthiness
Action Steps
- Identify potential vulnerabilities in RAG models using Unicode perturbations
- Analyze query and external code repositories for hidden UTF characters
- Develop and implement detection methods for RAG-Pull attacks
- Test and evaluate the effectiveness of RAG-Pull attack detection
- Apply security patches and updates to prevent RAG-Pull attacks
Who Needs to Know This
AI researchers, developers, and security experts can benefit from understanding RAG-Pull attacks to improve LLM security and robustness
Key Insight
💡 RAG-Pull attacks compromise LLM reliability and trustworthiness by injecting malicious code via invisible Unicode perturbations
Share This
🚨 New RAG-Pull attack injects malicious code into LLMs via invisible Unicode perturbations 🚨
Key Takeaways
Learn how RAG-Pull attacks inject malicious code into LLMs via invisible Unicode perturbations, compromising model reliability and trustworthiness
Full Article
Title: RAG-Pull: Turning Retrieval into a Code-Injection Channel via Invisible Unicode Perturbations
Abstract:
arXiv:2510.11195v2 Announce Type: replace-cross Abstract: Retrieval-Augmented Generation (RAG) increases the reliability and trustworthiness of the LLM response and reduces hallucination by eliminating the need for model retraining. It does so by adding external data into the LLM's context. We develop a new class of black-box attack, RAG-Pull, that inserts hidden UTF characters into queries or external code repositories, redirecting retrieval toward malicious code, thereby breaking the models' s
Abstract:
arXiv:2510.11195v2 Announce Type: replace-cross Abstract: Retrieval-Augmented Generation (RAG) increases the reliability and trustworthiness of the LLM response and reduces hallucination by eliminating the need for model retraining. It does so by adding external data into the LLM's context. We develop a new class of black-box attack, RAG-Pull, that inserts hidden UTF characters into queries or external code repositories, redirecting retrieval toward malicious code, thereby breaking the models' s
DeepCamp AI