AI Agent Security (OWASP Agentic Top 10) Fundamentals Training
A one-day fundamentals programme on OWASP Agentic Top 10. Learn the main security risks when AI agents can use tools, make decisions and perform tasks independently. Participants work through practical examples and a guided case before drafting actions for their own organisation.
Who should attend: AI product owners, developers, security reviewers, IT and automation teams. No prior framework knowledge is required.
HRD Corp claim support available
Programme Agenda
9:00 AM to 10:00 AM
What Changes When an Agent Acts
Map an agent's goals, tools, identities, memory and human approvals. Introduce the OWASP Top 10 for Agentic Applications 2026 and distinguish agent risks from a chatbot's output risks.
10:00 AM to 10:15 AM
Break
10:15 AM to 11:30 AM
Goals, Tools and Authority
Explore goal hijacking, tool misuse and identity or privilege abuse. Use a refund agent to identify who authorised an action, what it can access and where a restriction should be enforced.
11:30 AM to 12:30 PM
Code and Supply Chain Exposure
Recognise insecure dependencies, unsafe execution paths and compromised agent components. Discuss approved tools, restricted execution and supplier checks without running an exploitation lab.
12:30 PM to 1:30 PM
Lunch
1:30 PM to 2:45 PM
Memory and Agent Communications
Consider untrusted information stored in memory and exchanged between agents. Identify provenance, validation and isolation questions when one agent's output becomes another agent's instruction.
2:45 PM to 3:00 PM
Break
3:00 PM to 4:00 PM
Failures, Human Trust and Recovery
Explore cascading failures, misleading explanations and loss of control. Define human checkpoints, monitoring, action limits and ways to pause an agent or recover from an incorrect operation.
4:00 PM to 5:00 PM
Practical Exercise: A Refund Agent Review
Assess a fictional agent that reads messages and issues refunds. Identify risks across the current Top 10, select safeguards and draft release conditions and evidence requirements.
Key Outcomes
- Identify major agentic application risks
- Map tool and privilege boundaries
- Choose oversight and containment measures
- Draft a basic agent security review
Training Mode Physical / Online / Hybrid
HRD Corp Subject to programme eligibility and HRD Corp approval
Level Fundamentals / Beginner. No prior framework knowledge required.
Duration 1 Day | 9:00 AM to 5:00 PM (6.5 learning hours, excluding breaks)
Venue In-house at the client's premises, or delivered via the client's preferred platform (Microsoft Teams, Zoom, or equivalent)
Assessment Group case exercise and a 10-question knowledge check
Certificate Orbix Certificate of Completion upon full attendance; not a professional certification
Malaysia Relevance
High relevance for Malaysian automation providers and organisations connecting AI agents to business systems. The 2026 agentic application guidance supports risk identification; it is not a product certification.
Practical Takeaways
An agent trust-boundary map and risk-to-safeguard review checklist.
Download a sample practice worksheet (TXT). This fictional exercise is a preview of the learning approach. About Orbix; contact us to confirm the facilitator and relevant experience for your delivery.
Course Completion and Next Steps
Participants receive an Orbix Certificate of Completion upon full attendance. This introductory course does not confer professional certification or establish an organisation's compliance. Further implementation and assessment depend on the organisation's scope and obligations.
Frequently Asked Questions
Related Fundamentals Training
AI Security Verification (OWASP AISVS) Fundamentals Generative AI Security (OWASP LLM Top 10) Fundamentals AI Agent Runtime Security (OWASP Agent Control Standard) Fundamentals AI Security Controls (CSA AICM) Fundamentals View all fundamentals programmes
Reference material: OWASP: Top 10 for Agentic Applications 2026.