This course contains the use of artificial intelligence.

93 tools for security testing of AI agents in one short course: 78 ready-to-use record templates, checklists and test procedures, 8 downloadable tool files and 7 models.

Open a lecture, take the file, run the test in the supplied sandbox.

Why prompt testing is not enough for agents

Traditional prompt testing often ends when the model returns a response. An AI agent can fail later: after a retrieved document changes its goal, after an allowed tool result feeds another call, after a poisoned memory record returns in a new session, or after one peer passes tainted authority to another. A final refusal can hide an earlier planning deviation, and a successful tool block can contain impact without fixing the source. Effective agentic red-teaming follows the complete trajectory.

What changes after this course

You move from isolated prompts to bounded security experiments. You define protected properties before each run, compare clean and adversarial conditions from the same frozen state, keep the causal links between events, and verify both containment and retained utility. You can name the first changed field, the control that allowed or blocked it, the downstream components affected and the final simulated state.

Safe by design

Every exercise uses a fictional security operations workflow and a synthetic incident. The environment contains mock identities, fake canaries, local sinks, reversible state, deterministic scripts and explicit stop conditions. No exercise requires production credentials, customer data, public targets, destructive payloads or third-party systems. The Sandboxed Agent Lab is a self-contained browser file with no external network, no model call and no live integration.

What is inside: 93 tools, counted lecture by lecture

You do not have to take this number on trust. Open any lecture and read its description: every tool inside is listed there by name.

• 78 ready-to-use tools inside the presentations:
• 44 templates and records: agent attack surface map, agent goal integrity baseline card, agent tool call evidence record, red-team finding template
• 14 checklists: agent memory provenance, agent refusal and recovery, agent tool-chain containment, red-team stop conditions
• 14 step-by-step procedures: clean agent baseline test, reversible agent sandbox execution, controlled agent memory injection, agent sandbox reset verification
• 6 other tools: agent failure metrics, agent release action-zone matrix, red-team coverage matrix, agent owner attack surface interview
• 8 downloadable tool files, one for every lecture: Agent Attack Surface Map, Multi-Turn Goal Hijack Test Pack, Tool Affordance Chain Mapper, Memory Poisoning Test Protocol, Multi-Agent Scenario Builder, Sandboxed Agent Lab, Agent Failure Metrics and Agentic Red-Team Engagement Pack
• 7 models explained with examples: agent loop, trust boundary mapping, confused deputy, affordance chain analysis, circuit breaker

Where the tools are

• Agentic AI attack surfaces and goal hijacking (2 lectures, 20 tools).
• Tool security, memory poisoning and multi-agent attacks (3 lectures, 36 tools).
• AI agent sandboxing and failure cascade metrics (2 lectures, 26 tools).
• Agentic red-team campaign and release decisions (14 tools).

Who is teaching

I am Mike, the #1 HR instructor on Udemy, with 1,800,000+ course enrollments from 170,000 students, and the author of about 200 books on Amazon. I built the people system at Preply as it became a unicorn and worked at Wargaming, iDeals and Alfa-Bank across 20 years in HR. I hold PHRi, SHRM-CP and HCI sHRBP certifications and represent HRCI in Eastern Europe.

I come to this topic as a person who designs and buys such systems. I once helped a startup that built a chatbot for employees and described about 50 such agents for them: what each one does, what it has access to and how it talks to the user. Developers then built them from my descriptions. I also know the other side of the table. At iDeals, a virtual data room company trusted by clients in mergers and acquisitions, the security requirements were very high. We negotiated for a long time with a platform that predicted engagement and turnover, and the question of which data it collected and where it sent them became the problem: we could not let any data of our employees leave. An agent that plans, remembers and calls tools raises the same question many times over, and this course gives a structured way to answer it with evidence.

How to use the course

Go through the eight lectures in order: each one adds a test and a file, and the last lecture joins them into one campaign. It takes seven hours. If you do not have an authorized company sandbox, the supplied case is a complete practice path. Section and lecture titles show how many tools are inside, and each lecture description lists them by name.

What is included

• 8 lectures across 4 sections, 7 hours
• 16 downloadable resources: 8 tool files and the lecture presentations
• 93 tools across the lessons, ready to use in the supplied sandbox
• Lifetime access to all materials
• Active instructor support in the Q&A section
• A Udemy Certificate of Completion

Where to start

Open the first lecture, download the Agent Attack Surface Map and fill it in for one agent you know: its goal, its tools, its memory and its peers. Enrol now and start your first lesson today.