From f85c91998988bdacb03a1c8be0228501e63eb93f Mon Sep 17 00:00:00 2001 From: Matteo Meucci Date: Sun, 16 Nov 2025 16:42:18 +0100 Subject: [PATCH] Update 2.1.1_Architectural_Mapping_of_OWASP_Threats.md --- .../content/2.1.1_Architectural_Mapping_of_OWASP_Threats.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/Document/content/2.1.1_Architectural_Mapping_of_OWASP_Threats.md b/Document/content/2.1.1_Architectural_Mapping_of_OWASP_Threats.md index c09a43f..97ce3a4 100644 --- a/Document/content/2.1.1_Architectural_Mapping_of_OWASP_Threats.md +++ b/Document/content/2.1.1_Architectural_Mapping_of_OWASP_Threats.md @@ -2,13 +2,13 @@ In this chapter, we present a structured mapping of AI security threats from the OWASP Top 10 LLM Risks (2025) and the OWASP AI Exchange Threats onto a modular AI system architecture, grounded in Google’s Secure AI Framework (SAIF). -By examining the AI architecture across its four core layers, data, infrastructure, model, and application, we can visually pinpoint where threats are most likely to materialize as risk exposure, thereby enabling focused and effective security testing. Figure 2.1, titled ‘OWASP AI Threats Mapped to AI Components,’ illustrates this alignment and serves as a reference for mapping threats to the specific components within the AI system. +By examining the AI architecture across its four core layers, data, infrastructure, model, and application, we can visually pinpoint where threats are most likely to materialize as risk exposure, thereby enabling focused and effective security testing. Figure 2, illustrates this alignment and serves as a reference for mapping threats to the specific components within the AI system.

AI Architecture Threat Model

-**Fig 2.1 Threat Model of OWASP Threats (LLM T10 and AI Exchange) Mapped to Impacted AI Components of a SAIF baseline architecture** +**Fig 2 Threat Model of OWASP Threats (LLM T10 and AI Exchange) Mapped to Impacted AI Components of a SAIF baseline architecture** We use a structured process for identifying potential threats to an AI system by analyzing its architecture and operational context. In this approach, we reference threat categories defined by OWASP, specifically the *OWASP Top 10 for LLM* and *OWASP AI Exchange* to identify risks such as prompt injection, data poisoning, and model evasion. For each identified threat, we outline a representative threat scenario to highlight which system components are impacted. This mapping helps derive targeted test cases aimed at uncovering exploitable vulnerabilities and weaknesses.