Securing Agentic AI: From AI Security Fundamentals to Hands-on Agent Assessment — Antitech Workshop
THE JUDGES AT AABW AREN'T EVALUATING DEMOS. THEY'RE EVALUATING DEPLOYMENTS.
Every problem statement at this hackathon comes from a real enterprise. Every project is judged on production-readiness, not demo polish.
That means your agent will be evaluated against the same standard a security review would apply before a company greenlights a deployment: Can this thing be manipulated? Does it leak data? Can a user make it do something it shouldn't?
Most builds this week won't have an answer to those questions.
This workshop is how you get one — in 60 minutes.
🔴 WHAT THIS IS
A hands-on AI security workshop at VNG Campus, led by Achraf "Ash" Jday, Co-Founder of Antitech — building the security layer for the agentic age, backed by Antler, recognised as Best AI Solution in Japan at The Ventures 2026.
In 60 minutes, you will:
Map the 8 attack surfaces enterprise security teams look for first in agentic systems: prompt injection, indirect prompt injection, conversation-led attacks, memory poisoning, unsafe instruction following, data leakage, tool abuse, and policy bypass — and understand why your existing security thinking doesn't catch any of them
Run real red-team checks using the Antitech SDK against actual agent types: chatbots, workflow agents, API-connected agents, document-ingestion agents, and enterprise-style agents
Apply blue-team hardening in the same session: PII/data obfuscation, anti-prompt-injection defenses, safer input/output handling — patterns you can implement on your AABW build tonight
Generate a structured Agent Assessment Report — the exact format enterprise evaluators use: vulnerabilities found, severity, affected components, reproduction steps, mitigations applied
Optional deliverable: Teams that run the SDK assessment during Build Week can submit an Agent Assessment Report alongside their project. It documents what you found, what you fixed, and why your agent is deployment-ready. It's the kind of evidence most teams won't have. It's the kind judges notice.
🔬 HOW THE SESSION IS STRUCTURED
AI Security Foundations — 25–50 min
The agentic AI risk landscape: why traditional app security — authentication, validation, rate limiting — is not enough once your agent reads context, executes tools, and chains decisions autonomously. What goes wrong, where, and why.
Antitech SDK Walkthrough — 30–45 min
How to use the SDK to run red-team style assessments: test for prompt injection, memory poisoning, data leakage, tool abuse, and policy bypass. Structure your findings. Generate a report with severity ratings, reproduction steps, and remediation recommendations.
Use Cases + Q&A — remaining time
Concrete scenarios for the builds happening at AABW this week. How to apply this to your own agent before submission closes.
👥 WHO THIS IS FOR
This is for you if:
You're building an agent at AABW and want it to survive enterprise evaluation — not just Demo Day
You haven't seriously asked "what happens when someone tries to break my agent?" — and you know you should have
You want a submittable security artifact that demonstrates production-ready thinking to judges
This is NOT for you if you want a general AI ethics overview. This is offensive and defensive security engineering — hands-on, for builders who are shipping real things under real constraints.
🎯 YOUR SPEAKER
Achraf "Ash" Jday — Co-Founder, Antitech. Backed by Antler. Best AI Solution in Japan, The Ventures 2026.
Before Antitech, Achraf designed hundreds of infrastructure architectures, governance frameworks, and agentic interfaces for enterprise clients in banking, industrial, and space tech. He's trained and mentored thousands of builders worldwide through programs connected to Microsoft, MIT, Stanford, and EPFL.
Sorbonne MSc AI & Cybersecurity — 21/20 thesis score on AI transformation in fintech security. NASA Space Apps France winner. WorldSkills Île-de-France Cybersecurity Champion. EPFL LauzHack winner. Top 5 in Facebook's Cybersecurity Challenge as a teenager — the youngest competitor in the field. France's Lead Microsoft MLSA Ambassador.
His mission: make the future's autonomous systems harder to hack than they are to build.
⚠️ ONE REQUIREMENT
You must be registered for Agentic AI Build Week to attend. Every application is reviewed — builders actively working on something at AABW get priority.
Haven't applied yet? Register for the main event first.
THE CONTEXT
This session is part of Agentic AI Build Week (Jul 8–12) — Southeast Asia's largest agentic AI hackathon, with 2,000+ builders, 30+ tech partners, and a $1M+ prize pool across Ho Chi Minh City.
👉 Apply below. Build something that doesn't just demo well — build something an enterprise would actually deploy.