Resources
AI Agent Security Guides, Insights & Examples
Use these guides and examples to prepare requirements, tests and review evidence.
Start here
Security review checklist
Questions and artefacts to prepare before a review.
OWASP agentic risk guide
Turn risk categories into scoped control questions.
Sample assessment report
Fictional findings, evidence and acceptance criteria.
Action-control demonstration
A runnable local example of argument checks.
Explore the collection
Ideas for the
work ahead.
Find a practical perspective on the controls, questions and evidence that matter to your deployment.
11 resources
OWASP Agentic Top 10 Security Review Guide
Turn OWASP agentic risk categories into scoped control questions and review evidence. Use the taxonomy as a starting point, not a certification or full test suite.
Read the perspectiveAI Agent Security Review Readiness Checklist
Prepare system ownership, agent inventories, control tests and evidence before a buyer or internal security review. Record gaps, owners and acceptance tests.
Read the perspectiveAI Agent Guardrails at the Action Boundary
Understand why model intent is not authorisation. Enforce tool permissions, resource limits and approval conditions at the AI agent's execution boundary.
Read the perspectiveEU AI Act Article 15: Technical Evidence Planning
Establish AI Act applicability before selecting Article 15 technical evidence. Link cybersecurity controls, test scope and results to the system under review.
Read the perspectiveHuman Approvals for AI Agents Without Fatigue
Design useful human approval requests for AI agents with action, target and consequence context. Test changes, expiry and exceptions before rollout.
Read the perspectiveMCP Tool Pinning, Changes & the Rug-Pull Risk
Review MCP tool owners, versions and permissions when behaviour changes. Understand the limits of pinning and rerun relevant access and action tests.
Read the perspectiveThird-Party AI Agent Risk: Review the Deployment
Review supplier AI agents against the actions and access enabled in your deployment. General supplier security evidence may not cover those permissions.
Read the perspectiveAI Guardrail Benchmarks: Read the Test Design
Interpret AI guardrail results using test design, scenario coverage and observed outcomes. Avoid treating one headline score as a readiness prediction.
Read the perspectiveAI-Assisted Cyber Operations: Control Scenario
Explore how AI-assisted reconnaissance and tool use can connect attack attempts. Use this generalised scenario to review permissions and monitoring boundaries.
Read the perspectiveMCP Tool Poisoning: Threat Scenario & Controls
Understand how malicious MCP tool descriptions or output can influence an agent. Review tool trust, change detection and execution boundary controls.
Read the perspectivePrompt Injection & Private Data Leakage Risks
Review how untrusted content can redirect agent retrieval and exports. Test data boundaries, permitted destinations and action controls with synthetic data.
Read the perspectiveNo matching resources.
Try a broader term or choose another category.
Turn security challenges into a clear plan
Tell us what you need to protect, improve or achieve. We will help you identify the control gaps, prioritise the work and define practical steps towards your goal.