Learn AI security
Clear guides on how applications built on language models are attacked and defended, and which AI Labs challenges let you practise each technique.
How to get started in AI security with a CTF
An AI CTF is a hands-on way to get started in AI security: you learn by attacking systems built for that purpose. This guide covers useful background, the first concepts to learn, how the competition works and a path you can start for free.
How to deploy and hack a boot2root machine
This guide covers how to set up the connection, how to deploy a machine on AI Labs and which method takes you from the first scan to root, with the AI component as the way in.
OWASP Top 10 for LLM Applications (2026) explained
The OWASP Top 10 for LLM Applications is the taxonomy AI Labs uses to classify its challenges. This guide covers every risk in the 2026 edition with a generic example and a defense idea, plus what changed from 2025.
Prompt injection explained: direct and indirect attacks (OWASP LLM01)
Prompt injection is the top risk in the OWASP Top 10 for LLM Applications. This guide explains why it works, how direct and indirect prompt injection differ and which controls reduce the risk in a real application.
Sensitive information disclosure and system prompt leakage in LLM applications
LLM data leakage ranges from exposing personal data to revealing the system prompt and the rest of the hidden context. Learn why it happens, where it shows up in real integrations and how to prevent it.
AI agents and excessive agency: when the model takes action
Learn what an AI agent is, how it calls tools and MCP servers, why excessive agency ranks among the main risks in the OWASP Top 10 for LLM Applications, and how to limit what an agent can do.
RAG poisoning: attacking retrieval, embeddings and assistant memory
Learn how retrieval-augmented generation (RAG) works, how attacker-controlled data reaches the model's context, and which controls prevent RAG poisoning and memory poisoning in AI assistants.
Improper output handling in LLM applications: XSS, SQL injection and command execution
Improper output handling happens when an application trusts what the model generates and passes it unchecked to a browser, a database or an interpreter. Learn how attackers exploit it and which controls stop it.
AI supply chain security: models, dependencies and MCP servers
Learn which third-party components make up the supply chain of an AI application, how each one gets compromised, and which controls reduce AI supply chain risk before you load a model or connect a tool.
Unbounded consumption in LLM applications (OWASP LLM06)
Every request to a language model costs compute and money. This guide explains how an attacker turns that bill into a weapon, which other threats fall under the category and which controls keep it in check.
Misinformation and hallucinations in LLM applications (OWASP LLM07)
A language model can give a false answer with the same confidence as a true one. This guide explains why that happens, when it becomes a security problem and which controls reduce it.