AI Security Engineer & Researcher LLM Security • AI Agent Security • AI Red Teaming • Prompt Injection

I build AI systems, study how they fail under adversarial conditions, and design security controls to make them safer to deploy.

My work focuses on the security boundary between LLMs, agents, tools, data, memory, and external systems.

Build it. Break it. Understand it. Secure it.

• LLM Security — Prompt Injection, Jailbreaking, System Prompt Leakage
• AI Agent Security — Tool Abuse, Excessive Agency, Memory Poisoning, Privilege Abuse
• AI Red Teaming — Adversarial testing, attack simulation, security assessments
• RAG Security — Data Poisoning, Retrieval Manipulation, Context Injection
• Tool & API Security — Unauthorized Tool Calls, Data Exfiltration, Trust Boundary Failures
• Deep Learning Security — Adversarial ML, Model Robustness & Security Research
• AI Threat Modeling — OWASP, MITRE ATLAS, Attack Surface Analysis

I investigate how AI systems behave when their assumptions are deliberately violated.

Current research areas:

• Indirect Prompt Injection
• Agent Tool Hijacking
• Memory Poisoning
• AI Agent Privilege Escalation
• Data Exfiltration through Tool Calls
• RAG Poisoning
• Multi-Agent Security
• MCP Security
• Runtime Security for Autonomous Agents

My goal is to move beyond demonstrations and produce reproducible attacks, measurable benchmarks, and practical mitigations.

A collection of AI security assessments, adversarial scenarios, penetration-testing writeups and technical research.

Focus: LLMs • Agents • Prompt Injection • Data Exfiltration • Tool Abuse

An experimental laboratory for studying the security of neural networks and machine-learning systems.

Focus: PyTorch • Neural Networks • Adversarial ML • Model Robustness

An experimental runtime security layer for AI agents.

Focus: Prompt Injection Detection • Tool Call Analysis • Policy Enforcement • Agent Security

I am progressively moving my work toward reproducible security research:

Attack → Exploit → Measurement → Mitigation → Benchmark

Where possible, research includes:

• reproducible Proofs of Concept
• attack datasets
• test environments
• measurable attack success rates
• mitigation experiments
• technical writeups

My work references and maps research against established security frameworks including:

• OWASP Top 10 for LLM Applications
• OWASP Top 10 for Agentic Applications
• OWASP Agentic AI Threats & Mitigations
• MITRE ATLAS
• CVSS

AI / ML

Python PyTorch Transformers Neural Networks LLMs

AI Engineering

LangChain CrewAI APIs RAG Multi-Agent Systems

Security

AI Red Teaming Prompt Injection Web Security OSINT Pentesting

Security Research

Threat Modeling Adversarial Testing Vulnerability Research Security Automation

I publish technical research and practical security analysis covering the emerging attack surface of AI systems.

Research topics: LLM Security • Agentic AI • Prompt Injection • Tool Security • AI Red Teaming

Founder of Vanguard AI & SECURITY and an AI Security Engineer focused on building and securing intelligent systems.

I am particularly interested in the intersection of:

Artificial Intelligence × Cybersecurity × Autonomous Agents

Email: antonio.redteam1@gmail.com

LinkedIn: Antonio Toudji

Website: https://linktr.ee/Antniotdj

Security research published here is intended for authorized testing, defensive research, education, and responsible disclosure.