AI Security

AI security means protecting AI systems from attack and misuse, and protecting your organisation from the new risks AI brings. Here are the main threats and…

Prompt injection tricks an AI system into ignoring its instructions. Here’s how direct and indirect prompt injection work, real-world scenarios, and how to reduce the risk.

The OWASP Top 10 for LLM Applications lists the most critical security risks in apps built on large language models. Here’s each risk in plain English,…

Sensitive data leaks through AI chatbots in two directions: staff pasting it in, and chatbots giving it out. Here are the common leak paths and practical…

A jailbreak tricks a large language model into ignoring its safety rules. Here’s how jailbreaks work, how they differ from prompt injection, and how to limit…

Data poisoning manipulates the data an AI model learns from, to plant backdoors, biases or errors. Here’s how it works, where it happens in practice and…

Most AI is built from third-party parts: pre-trained models, datasets, libraries, plugins and APIs. Here are the supply chain risks in each layer and how to…

AI agents don’t just answer questions. They read email, browse, call APIs and take actions. Here are the security risks that creates and how to deploy…

Shadow AI is staff using AI tools the organisation hasn’t approved or doesn’t know about. Here’s why it happens, the risks it creates and how to…

Red teaming tests your AI chatbot the way a determined user or attacker would, before launch. Here’s a practical, beginner-friendly process: scope, test areas, scoring and…