Protecting AI systems and the organisations that use them: attacks, defences and secure practices.
Using language models and machine learning inside a security team without creating a new incident class.
Attacks on models themselves: evasion, poisoning and backdoors, model theft, and what the data remembers.
Access control across a retrieval index, the confused deputy problem in tool use, and keeping an agent inside its blast radius.
How injection works, why filtering fails, and the design patterns that actually contain it.
What actually changes when a model joins a system, the attacks that follow from it, and how to threat model an AI feature before you ship it.
35 in this topic
What AI security covers — attacks on models, data and AI applications — and how it differs from traditional security.
AI security 1 min read 7 Jul 2025 For people
An overview of the widely used list of the most critical security risks for applications built on language models.
AI security 1 min read 6 Jul 2025 For people
How people try to get models to bypass their safety training, common techniques, and layered defences.
AI security 1 min read 5 Jul 2025 For people
How attackers hide instructions in web pages, emails and documents that AI systems read, and why it's so dangerous for agents.
AI security 1 min read 4 Jul 2025 For people
How attackers corrupt training or fine-tuning data to change model behaviour, and how to protect data pipelines.
AI security 1 min read 3 Jul 2025 For people
How hidden triggers can be planted in models, why they're hard to detect, and how to reduce the risk.
AI security 1 min read 2 Jul 2025 For people
How tiny, carefully crafted changes to inputs can fool machine learning models, and approaches to robustness.
AI security 1 min read 1 Jul 2025 For people
How attackers steal models by copying weights or imitating them through queries, and how to protect model assets.
AI security 1 min read 30 Jun 2025 For people
How models can reveal information about their training data — membership inference, memorisation, inversion — and defences.
AI security 1 min read 29 Jun 2025 For people
Practical controls for agents that use tools and act autonomously: least privilege, isolation, approval and monitoring.
AI security 1 min read 28 Jun 2025 For people
Why model output must be treated as untrusted input to other systems, and how to avoid injection and unsafe actions.
AI security 1 min read 27 Jun 2025 For people
Keeping API keys, passwords and tokens out of prompts, training data and model reach.
AI security 1 min read 26 Jun 2025 For people
Managing the risks of third-party models, datasets, libraries and tools in AI systems.
AI security 1 min read 25 Jun 2025 For people
Protecting the servers, pipelines and files that train and serve AI models.
AI security 1 min read 24 Jun 2025 For people
How AI-generated audio, video and images are used in fraud and disinformation, and how organisations can defend.
AI security 1 min read 23 Jun 2025 For people
How attackers use AI to scale phishing, find vulnerabilities and automate attacks, and what it means for defenders.
AI security 1 min read 22 Jun 2025 For people
A structured way to identify how an AI system could be attacked or misused before building defences.
AI security 1 min read 21 Jun 2025 For people
Planning and running adversarial security tests against AI applications, and turning findings into fixes.
AI security 1 min read 20 Jun 2025 For people
What to log and watch in production AI applications to detect attacks, misuse and leaks.
AI security 1 min read 19 Jun 2025 For people
Preparing for and handling security incidents involving AI applications, models and agents.
AI security 1 min read 18 Jun 2025 For people