Runtime Security for AI Agents: Monitoring What Agents Actually Do
Preventive controls for AI agents will never be perfect, because prompt injection has no complete fix. Runtime security fills the gap: captu…
Category
AI Security
30 articles tagged AI Security.
AI-Driven Vulnerability Discovery: Are We Entering a Vulnerability Explosion?
AI systems can now find real, exploitable bugs in mature code, and they are being pointed at everything. That is good news for software qual…
AI Agents as Non-Human Identities: A New IAM Security Problem
AI agents act on behalf of people, teams and other agents, which makes them non-human identities with an unusual twist: delegated, dynamic a…
MCP Security: The New Security Boundary for AI Applications
MCP has become the standard way AI applications reach enterprise systems, and that makes the MCP layer a security boundary in its own right,…
The Agentic Attack Surface: Identity, Tools, Memory and Autonomous Actions
A tester's view of agentic systems. We break the agent attack surface into four areas (identity, tools, memory and autonomy), describe the a…
The Agentic Attack Surface: Securing Autonomous AI Systems
A year ago, most AI agents were pilots. Now they run inside ticketing systems, CI pipelines, SOCs and customer service, often with more acce…
AI Red Teaming for Developers: Testing Agents Before Production
Red teaming used to be something a security team did to a finished system. For AI agents, it has to start with the developers who build them…
AI-Powered Cyberattacks: How Attackers Are Automating Reconnaissance and Exploitation
For two years, 'AI-powered attacks' mostly meant better phishing emails. This summer, public reports showed criminals using agentic AI tools…
MCP Security: Securing the Connection Between AI Agents and Tools
A practical hardening guide for security teams responsible for MCP deployments: how to assess MCP servers, configure clients, apply the prot…
MCP Security: What Happens When LLMs Can Call Your APIs?
The Model Context Protocol has become the standard way to connect AI assistants to tools and data in barely six months. Security researchers…
Securing AI Agents: A New Cybersecurity Attack Surface
AI agents that browse, write code and operate software on a user's behalf moved from demos to products in the last few months. They combine…
OWASP for LLMs and Agents: Understanding the New AI Attack Surface
OWASP's GenAI Security Project has published a refreshed 2025 Top 10 for LLM Applications and, last month, a dedicated threat taxonomy for a…
Agentic AI Security: When AI Can Actually Take Actions
Computer-use and browser agents can now click, type and navigate like a person, in real browsers and desktops, on your behalf. That turns ev…
From SAST to AI Security Testing: What's Next for Application Security?
In November, Google reported an AI agent finding an exploitable bug in SQLite before release, and AI-generated fuzz targets uncovering vulne…
AI Security Testing: How to Red-Team an LLM Application
Most organisations now have LLM features in production, and many have never tested them adversarially. This is a practical methodology for r…
Securing RAG Applications: Where Retrieval Becomes an Attack Surface
Retrieval-augmented generation is how most organisations connect LLMs to their own data, and it introduces risks that are easy to miss: assi…
LLM Supply-Chain Security: Models, Datasets, Embeddings and Dependencies
Last week, researchers reported around a hundred malicious models on Hugging Face that could execute code on anyone who loaded them. AI appl…
AI Security: What Changes When Your Application Starts Using LLMs?
Adding a large language model to an application changes its threat model in ways traditional AppSec does not cover. Input becomes instructio…
AI Red Teaming: How Security Testing Changes With LLMs
Thousands of hackers spent DEF CON trying to break AI models, the White House has secured commitments from AI companies to red-team their sy…
LLM Security Explained: How to Attack and Defend AI Applications
OWASP published its first Top 10 for Large Language Model Applications on 1 August. Meanwhile, researchers have shown automated jailbreaks t…
Prompt Injection: The First Major Security Problem of LLM Applications
Within days of launch, users persuaded Bing Chat to reveal its hidden instructions, and researchers showed that web pages can silently take…
Using Large Language Models for Security Research and Threat Intelligence
Within days of ChatGPT's release, security researchers were using it to explain code, draft detection rules and summarise threat reports, an…
AI-Assisted Phishing: How Generative Models Change Social Engineering
Large language models can now write fluent, personalised emails in seconds, and researchers have shown AI-written spear-phishing can outperf…
Generative AI Meets Cybersecurity: What Happens When AI Can Generate Code?
GitHub Copilot became generally available in June, and code-generating models are now part of everyday development. That has consequences fo…
Deepfakes and Synthetic Media: The New Social Engineering Problem
Criminals have used AI-cloned voices to impersonate executives and authorise large transfers, and the FBI has warned that synthetic media wi…
AI-Powered Threat Detection: From SIEM Rules to Behavioral Analytics
SolarWinds showed how a patient attacker using legitimate credentials and tools can slip past rule-based detection for months. Behavioural a…
Adversarial Machine Learning: How Attackers Can Fool AI Security Systems
As security products rely more on machine learning, attackers are learning to target the models themselves. Tiny, carefully chosen changes c…
The Problem With Using AI for Security: False Positives, Bias and Adversarial Attacks
AI promises to help overwhelmed security teams, but it brings problems of its own: alert floods, models that learn the wrong lessons from bi…
Can Machine Learning Detect Malware Better Than Traditional Signatures?
Signature-based antivirus has been declared dead many times, and ML-based 'next-generation' endpoint protection has been sold as its replace…
AI in Cybersecurity: How Machine Learning Changed Threat Detection
Almost every security product now claims to use AI. Behind the marketing, machine learning has genuinely changed how threats are detected, f…