AI5 PROFESSIONAL CERTIFICATE · BY TGC INDIA · UPDATED AUGUST 2026
Agentic AI Testing & Safety
Test agent behaviour, permissions, output quality, failure paths and approval rules.
Online + OfflineWeekdays + WeekendsRegular + Fast Track
22+years in training
20,000+learners across TGC
5classroom locations
Live onlinejoin from anywhere
NEXT START OPTIONS
Choose a live batch.
COURSE OVERVIEW
What this course
is built to do.
Test AI agents for task success, tool permissions, prompt attacks, unsafe behaviour and failure recovery.
Each module combines a live trainer demonstration, guided lab, assignment and review before the next stage.
CONNECTED CAREER DIRECTIONS
Agent Evaluation EngineerAI Safety TesterAgentic AI QA SpecialistModel Risk Associate
IS THIS COURSE RIGHT FOR YOU?
Choose it for the right reason.
You already build AI agents
You need release evidence and permission tests
You want attack, failure and monitoring practice
YOUR LEARNING ARC
From guided foundation to finished work.
Scope
Define behaviour, risks and release gates.
Test
Check tasks, tools, state and completion.
Attack
Challenge prompts, data and permissions.
Control
Add guardrails, monitoring and rollback.
INDUSTRY TASKS
Practise the work, not only the tool.
- Create an agent behaviour test set
- Run prompt-injection and tool-permission tests
- Prepare a monitored release decision
PRO
LED
LED
YOUR TRAINING TEAM
Learn from experienced working professionals.
Live demonstrations, guided practice and direct project feedback are part of the course. Trainers update examples and tool coverage as professional practice changes.
WHAT YOU WILL BE ABLE TO DO
Course outcomes
01
Plan agent evaluations
A risk, behaviour and release-gate charter.
02
Test tasks and tool use
A scored suite covering state, tools and completion.
03
Find permission and attack failures
An adversarial test record with repaired cases.
04
Operate agents safely
A release dossier with monitoring and rollback steps.
TAKE THE NEXT STEP
Need fees, syllabus or the right batch?
An AI5 Academy advisor can help you compare mode, schedule and starting level.
DETAILED COURSE FLOW
5 learning modules
Test AI agents for task success, permissions, prompt attacks, tool misuse and failure recovery before release.
Agent risk and evaluation design
Define expected behaviour, prohibited actions and release evidence for an agent system.
CORE TOPICS
- Agent goals, tools and data access
- Normal and unsafe behaviour boundaries
- Task-success and safety measures
- Test ownership and release gates
GUIDED PRACTICE
Create a risk and evaluation charter for a selected agent.
MODULE DELIVERABLE
Agent test and risk plan
Task, tool and state evaluation
Check whether the agent completes work correctly across normal and difficult cases.
CORE TOPICS
- Task test sets and expected outcomes
- Tool selection and parameter checks
- State, memory and hand-off accuracy
- Loop, timeout and incomplete-work detection
GUIDED PRACTICE
Run a test suite and classify failures by prompt, model, tool or workflow cause.
MODULE DELIVERABLE
Agent task-evaluation report
Prompt attacks and permission testing
Challenge the agent without allowing it to exceed approved access.
CORE TOPICS
- Prompt injection and indirect attacks
- Data leakage and secret handling
- Tool permission and privilege tests
- Content and action restrictions
GUIDED PRACTICE
Build an adversarial test set and verify that restricted actions remain blocked.
MODULE DELIVERABLE
Agent security test suite
Guardrails, approval and recovery
Add controls that stop or redirect unsafe and uncertain behaviour.
CORE TOPICS
- Input and output validation
- Confidence and rule-based routing
- Human approval and escalation
- Retry, fallback and incident paths
GUIDED PRACTICE
Repair selected failures and confirm the controls through regression tests.
MODULE DELIVERABLE
Controlled agent release candidate
Monitoring and safety capstone
Track behaviour after launch and respond to changing tools, prompts and data.
CORE TOPICS
- Tracing, audit logs and alerts
- Cost, latency and behaviour drift
- Incident review and rollback
- Evaluation schedule and model cards
GUIDED PRACTICE
Present a tested agent with release decision, monitoring and rollback plan.
MODULE DELIVERABLE
Agentic AI testing capstone
HOW THE TRAINING WORKS
Learn it. Apply it. Get it reviewed. Improve it.
Every important skill moves through explanation, demonstration, guided use and independent application. Trainer feedback is used to revise the work before it becomes part of the final portfolio.
Concept briefing
The trainer explains the principle, use case, limitations and the quality standard expected.
Live demonstration
A complete task is demonstrated while the trainer explains decisions, checks and common mistakes.
Guided lab
Learners repeat the method with support, ask questions and correct problems during the session.
Applied assignment
The same method is used on a different brief so the learner must make independent decisions.
Review and revision
Work is checked against a rubric, revised after feedback and prepared for project presentation.
PROGRESS IS CHECKED THROUGHClass exercisesModule deliverablesProject reviewsFinal capstone presentation
PORTFOLIO WORK
Projects you can show
Guided practice brief
Plan, produce, test and present a finished piece with trainer feedback.
Individual application
Plan, produce, test and present a finished piece with trainer feedback.
Workflow build
Plan, produce, test and present a finished piece with trainer feedback.
Industry-style assignment
Plan, produce, test and present a finished piece with trainer feedback.
Quality review
Plan, produce, test and present a finished piece with trainer feedback.
Final capstone
Plan, produce, test and present a finished piece with trainer feedback.
TAKE THE NEXT STEP
Need fees, syllabus or the right batch?
An AI5 Academy advisor can help you compare mode, schedule and starting level.
TOOLS COVERED
PythonEvalsLangSmithGuardrails
Tool coverage may be updated when the industry changes. Core methods remain part of the course.COMMON QUESTIONS
Before you apply
Is coding required?
Yes. Learners should be able to inspect and test Python agent workflows.
＋Which evaluation tools are used?
Current evaluation and tracing libraries are used according to the project.
＋Does this include prompt injection?
Yes. Direct and indirect attacks, data leakage and tool permissions are tested.
＋Do I need coding experience?
No, unless the course level says otherwise. Your advisor will check the right starting level.
＋Are classes live or recorded?
Classes are trainer-led in the classroom or live online. Recordings may support revision but do not replace class.
＋Will I receive a certificate?
Yes. Course completion requires attendance, assignments and the final project.
＋Can working professionals join?
Yes. Weekday, weekend and selected fast-track schedules are available.
＋NEXT BATCH
Choose your course.
Choose your schedule.
Online or offline. Weekdays or weekends. Regular or fast track. Speak with an AI5 Academy advisor about the right starting level.