ai on Antgarsil Pageshttp://pages.dosil.es/categories/ai/Recent content in ai on Antgarsil PagesHugo -- gohugo.ioenSat, 13 Dec 2025 10:45:00 +0100Exploiting LLM Tool Use via Prompt Injectionhttp://pages.dosil.es/posts/prompt-injection/Sat, 13 Dec 2025 10:45:00 +0100http://pages.dosil.es/posts/prompt-injection/Large Language Models (LLMs) are increasingly integrated into applications through mechanisms like “Tool Use” or “Function Calling.” While these integrations enable powerful automation, they introduce a new class of vulnerabilities: Prompt Injection. This is analogous to traditional injection attacks (SQLi, SSTI) but occurs within the natural language context of the model. The Vulnerability Prompt injection occurs when user-supplied input is concatenated with the “System Prompt” without sufficient boundary separation. If the model fails to distinguish between developer instructions and user data, it may follow malicious instructions embedded within the user input.