AI powered security operations bootcamp
                     AI-powered Security 
Operations Bootcamp
How LLMs Are Exploited by Hackers
11th - 12th July 2026 10 AM - 2 PM (IST)
TikToken
PyRIT
8 CPEs Hands-On
OWASP LLM Top 10 Agentic AI
Why Attend?
AI security is reshaping how modern enterprises build, deploy, and defend intelligent systems, 
creating both powerful capabilities and new attack surfaces. This bootcamp is designed to bridge 
that gap by combining AI fundamentals with real-world security practices. It focuses on how LLMs 
and agentic systems work, where they fail, and how they are exploited in production 
environments. Through structured, hands-on learning, it builds practical skills in AI threat analysis, 
red teaming, and secure system design for real-world security operations.
What sets this training apart:
AI Security Lifecycle LLM & AI Attack Agentic AI Security

Coverage Techniques
Understanding how LLMs work, Exploring prompt injection, Securing and attacking multi-agent 
fail, and are secured across the full jailbreaks, poisoning attacks, and systems, memory poisoning, and 
AI system lifecycle system prompt exploitation tool misuse scenarios
Hands-On Red Team Labs OWASP & MITRE Mapping AI Red Team Automation
Executing real-world AI attack and Mapping AI threats to OWASP Using tools like PyRIT and Garak 
defense scenarios using LLM, Agentic AI, MCP Top 10, and for structured adversarial testing 
frameworks, models, and tooling MITRE ATLAS and evaluation workflows
Key Takeaways
Earn 8 CPE credits
Hands-on AI security lifecycle training
Real-world LLM attack simulations
OWASP LLM and Agentic AI mapping
MITRE ATLAS threat scenario coverage
End-to-end AI red teaming experience
www.infosectrain.com
Meet the Expert
Urvesh 6+ Years of Experience
DFIR, Threat Hunting & Intel | CHFI | eTHP | DCPLA | CTIA | ECIH | 
CND | CCSE
Urvesh brings 6+ years of experience in information security, specializing in SOC operations, SIEM/
XDR implementation, detection engineering, threat hunting, and DFIR. He has worked extensively 
on building and managing Wazuh-based SOC environments, integrating threat intelligence, and 
automating incident response workflows.
His specializations include:
SOC deployment and SIEM/XDR implementation (Wazuh)

Detection engineering, custom rules, and alert tuning

Threat hunting and intelligence-driven investigations

SOAR-based incident response and automation

DFIR and security monitoring across hybrid environments
Urvesh has trained 300+ professionals in security operations, helping them build practical skills in 
detection, response, and SOC engineering.
www.infosectrain.com
Bootcamp Agenda
Day 1 
Module 1: The AI Security Landscape
Builds the foundational vocabulary of AI security - how LLMs actually work under the hood, where they break, and 
how to map AI threats to the frameworks SOC teams already use every day. 

How LLMs actually work: tokens, context windows, RLHF, temperature, sampling 

OWASP Top 10 for LLMs and MITRE ATLAS, mapped alongside MITRE ATT&CK

The AI attack surface from training data to inference and output handling 

Responsible AI security research and disclosure
Module 2: AI-Powered Threat Intel
Build a production-grade threat intel workflow where AI accelerates analyst work without becoming the source of 
truth & paste a real threat report URL, get back an analyst-ready evidence pack. 

AI-augmented IOC extraction from public threat reports 

Dynamic MITRE ATT&CK mapping with hallucination validation against the live ATT&CK dataset 

ATT&CK Navigator layer + SOC hunting pack generation (SPL + KQL starters)

Analyst-in-the-loop validation: final / review / rejected separation
Module 3: AI for GRC 
Use AI to draft, challenge, and validate compliance artifacts across ISO 27001, NIST CSF 2.0, EU AI Act, NIST AI RMF, 
and ISO 42001 without losing audit defensibility. 

ISO 27001:2022 Statement of Applicability drafting at scale 

NIST CSF 2.0 maturity assessment and gap analysis 

AI Governance through EU AI Act + ISO 42001 + NIST AI RMF lens 

GRC validation gate to catch hallucinated framework references and unsupported claims
www.infosectrain.com
Module 4: Assisted Detection Engineering Using AI 
Bring AI into every stage of detection engineering, from drafting Sigma rules to validating them against telemetry, and 
then turn detection engineering back on AI systems themselves. 

AI-assisted Sigma rule generation from attacker behaviour 

AI-powered log triage, incident timeline construction, and IR report drafting

Detection validation against real Windows Security and Sysmon logs 

Sigma detection rules FOR AI agent telemetry - tool-call anomalies, exfiltration patterns, memory-write events
Day 2 
Module 5: LLM Architecture
Understand the LLM internals that an attacker actually exploits, such as tokenisation, system prompt boundaries, 
sampling, and trust failures, because every attack starts here. 

Tokenisation deep-dive: token smuggling, homoglyphs, encoding bypass 

System prompt trust boundary failures 

Attack reliability across temperature and sampling: measuring ASR, not vibes

Uncensored vs aligned open-weight models: behavioural comparison
Module 6: Prompt Injection & Jailbreaks
The most active LLM attack surface in 2026, direct, indirect, and multi-turn, all covered with a measured Attack 
Success Rate. 

Manual Crescendo + Auto Crescendo (attacker/target/judge model loop) 

Skeleton Key direct policy override 

Many-shot Jailbreak and Context Compliance Attack (CCA) 

EchoLeak reproduction - CVE-2025-32711 markdown image exfil chain end to-end 

PoisonedRAG - embedding-layer poisoning of a local FAISS / Chroma knowledge base 

Judge-model scoring, telemetry capture, and Attack Success Rate measurement 
www.infosectrain.com
Module 7:  Agentic Attacks & Red Teaming
Where the 2025–2026 attack frontier actually lives - multi-agent systems, MCP servers, memory poisoning, and 
automated red-team tooling, all mapped to OWASP Agentic Top 10 (ASI01–ASI10). 

Build a 3-agent system (Orchestrator + Email + File), break it with indirect  injection, harden it with 3 defences 

Persistent memory poisoning across sessions, with SIEM-style detection 

Garak vulnerability scanning + aligned vs uncensored model comparison 

MCP Tool Poisoning & Rug Pulls - CVE-2025-54136 (MCPoison) and CVE 2025-54135 (CurXecute). 

PyRIT Automated Red Team - Microsoft's Crescendo and Tree-of-Attacks with Pruning orchestrators against your 
own targets
www.infosectrain.com
Contact us
[email protected]
www.infosectrain.com