Posts

Showing posts from August, 2026

OWASP Top 10 for LLM Applications 2026 - What Changed?

What is the OWASP LLM Top 10? The OWASP Top 10 for LLM Applications is a security awareness and risk framework for applications that use large language models. The 2026 edition continues that work but significantly changes the ranking and scope of several categories. One important change is how OWASP approached the ranking. The 2026 project compared practitioner opinion with evidence from 7,714 real incidents , with 6,639 incidents containing enough information to classify. Practitioner voting still carried most of the weight, while incident data contributed the remaining quarter. Why this matters: OWASP found that what security practitioners fear and what appears in public incident data do not always match. Prompt Injection remained #1 despite not appearing in the incident-data top 10, while Misinformation moved upward because the evidence showed it occurring more frequently than practitioners expected. Source: OWASP Top 10 for LLM Applications 2026. 2026 ...

TryHackMe Walkthrough: Securing AI Systems

AI applications introduce security challenges that go beyond traditional web application vulnerabilities. From LLM-powered assistants and RAG pipelines to AI agents connected to internal APIs, every new integration creates additional attack surfaces that security professionals need to understand. In this TryHackMe room, Securing AI Systems , you'll take on the role of a security architect reviewing an AI-powered code review assistant before it goes into production. You'll explore how production AI systems are structured, identify critical components and trust boundaries, and investigate security risks using industry frameworks such as the OWASP LLM Top 10 (2025) , MITRE ATLAS , and NIST AI Risk Management Framework . Through practical scenarios and an interactive AI system audit, you'll learn how issues such as excessive agency, sensitive information disclosure, system prompt leakage, improper output handling, and unbounded consumption can affect real-world AI applications....

Shadow AI: The Hidden Security Risk in Your Organization

Image
Understanding the risks of unauthorized AI usage in organizations Artificial Intelligence is now an essential part of everyday business operations. Employees use AI-powered tools to write code, summarize documents, generate reports, analyze data, create presentations and automate repetitive tasks. These tools significantly improve productivity and help teams complete work faster. However, not every AI application used within an organization is officially approved or monitored by its security team. Employees may independently sign up for AI services, install browser extensions or connect external applications to workplace accounts without understanding the security implications. This creates a growing cybersecurity concern known as Shadow AI . Although employees often adopt these tools with good intentions, unauthorized AI usage can introduce data exposure, privacy, compliance and third-party security risks. 🤖 What Is Shadow AI? Shadow AI refers to the use of artific...

MCP Security: Understanding the OWASP MCP Top 10 Vulnerabilities

Introduction Artificial intelligence applications are increasingly moving beyond simple question-and-answer interactions. Modern AI assistants can access external tools, retrieve information from internal systems, interact with APIs, and perform actions on behalf of users. These capabilities make AI applications more useful, but they also introduce new security considerations. The Model Context Protocol (MCP) is designed to make these integrations easier by providing a standardized way for AI applications to communicate with external tools and data sources. However, connecting an AI model to external systems also creates additional trust boundaries and potential attack surfaces. A compromised MCP server, poorly configured permission, malicious tool description, or exposed credential could affect more than the AI conversation itself. Depending on the permissions granted to the integration, an attacker may be able to access sensitive information, manipulate tool behaviour, or t...

RAG Security Fundamentals: Securing Retrieval-Augmented Generation

Image
Retrieval-Augmented Generation (RAG) has become a popular architecture for building AI applications that can answer questions using private, internal, or up-to-date information. Instead of relying only on an LLM's training data, a RAG system retrieves relevant information from a knowledge source and provides it to the model as context. This improves accuracy but also introduces a new security boundary. What is RAG? A simplified RAG workflow looks like this: User  > Application >  Retriever > Knowledge Base > LLM > Response For example, an employee asks:  "What is our company's remote-work policy?" The application searches internal documents, retrieves the relevant policy, and passes it to the LLM to generate an answer.  Typical RAG components include: Data sources:  PDFs, documents, databases, websites, tickets, etc. Document processing:  parsing and chunking data. Embeddings:  converting content into vectors. Vector database:  stor...

OWASP Agent Memory Guard: Protecting AI Agents from Memory Poisoning

Image
Introduction AI agents are becoming more capable of handling multi-step tasks, interacting with external tools, and maintaining context across multiple conversations. One capability that makes these systems more useful is persistent memory . Instead of treating every interaction as a new conversation, an agent can retain information such as user preferences, previous instructions, task history, and relevant business context. This allows it to provide more consistent and personalized responses. However, persistent memory also introduces a new security challenge: memory poisoning . If an attacker manages to insert malicious or misleading information into an agent's memory, that information may influence its decisions during future interactions. Unlike an attack that affects only one conversation, a successful memory poisoning attack can have a lasting impact. This makes agent memory an important part of the AI security attack surface. Memory Poisoning and OWASP Agent Memory Guard ove...

TryHackMe Walkthrough: AI Forensics

This walkthrough explores how Artificial Intelligence and Machine Learning can be applied to Digital Forensics and Incident Response (DFIR), while also highlighting the limitations and risks of relying on AI during forensic investigations. Task 2 - The AI Forensics Landscape Task Overview This task explores how AI/ML can be applied to Digital Forensics and Incident Response (DFIR) . It focuses on how AI can process large volumes of forensic data, detect anomalies, scale analysis across modern environments, and assist with tasks such as phishing detection, malware classification, alert triage, and timeline reconstruction. It also covers important AI limitations, including probabilistic behaviour, accuracy vs. precision and recall, and the "Garbage In, Garbage Out" principle . Questions & Answers 1. What ability of AI helps turn a DFIR investigator by recognizing patterns they might not have been able to comprehend? [REDACTED] 2. Which metric tells you the proportion of pos...

NIST AI Risk Management Framework: A Practical Guide to AI Security

Artificial Intelligence is becoming part of almost every modern organization. From chatbots and AI assistants to automated decision-making and security tools, organizations are rapidly adopting AI. But AI also introduces new security, privacy, and governance risks. To help organizations manage these risks, the National Institute of Standards and Technology (NIST) published the AI Risk Management Framework (AI RMF) The framework provides a practical approach for organizations to identify, assess, and manage risks associated with AI systems. What is the NIST AI RMF? The NIST AI RMF is a voluntary framework designed to help organizations build and use AI systems that are more trustworthy and responsible. Rather than focusing only on technical vulnerabilities, it considers risks across the entire AI lifecycle. The framework is built around four core functions: GOVERN  MAP  MEASURE  MANAGE These functions work together continuously rather than as a one-time process. 1. Govern...