Mohammad Raouf Abedini
AI Security Research · LLM Agent Red-Teaming · Offensive & Defensive Security Engineering
Security researcher focused on measuring and containing the cyber capabilities of frontier AI, ready to relocate now. The work is public, signed, and reproducible: an open-source containment-attestation framework, five DOI-archived Zenodo preprints (indexed under ORCID), and evaluation work inside Anthropic's safety-evaluation program.
Creator of Project Simurgh, a provider-agnostic containment-attestation framework that red-teams LLM agents under an adversarial, dishonest-producer threat model and produces Ed25519-signed, offline-verifiable evidence of what an agent did after a guardrail miss: 138/138 classifier-missed cases contained against a real Llama Guard 4, a live agent's attack success cut from 9/140 to 0/140 on AgentDojo, and five machine-checked Lean theorems. AGPL-3.0, 3,057 tests.
Also lead author of The Invisible Window (100% cross-platform screen-capture evasion, responsibly disclosed under OWASP/FIRST/CISA guidance) and Aion-BibleQA (a citation-faithfulness benchmark for retrieval-augmented LLMs: R@5 0.941, zero unsupported citations). Evaluated Claude outputs for exploitable code and guardrail circumvention in Anthropic's safety-evaluation program (via Alignerr, 2026). Cofounder of an incubator-backed campus-AI startup. Focused on reducing catastrophic risks from advanced AI by measuring capability uplift, characterising safety boundaries, and shipping defensive tooling.
Specializations
Vulnerability Research & Disclosure
- Cross-Platform Exploit Development (Win32 API, macOS ScreenCaptureKit)
- Responsible Disclosure (OWASP/FIRST/CISA Frameworks)
- IEEE-Format Security Research
- Penetration Testing & Secure Code Review
- W3C Screen Capture Specification
AI Safety & LLM Security
- LLM Integration & Evaluation (Claude, via Alignerr)
- AI-Assisted Vulnerability Research
- AI Capability Uplift Measurement
- Dual-Use Risk Assessment
- Intent-vs-Artefact Safety Boundary Characterisation
Python & Systems Programming
- Python (Primary), C, C++, TypeScript, Swift
- High-Performance C/C++ Systems (1M+ ops/sec)
- Network Packet Processing (libpcap)
- Linux (Ubuntu/Kali), CMake, Docker
- GitHub Actions CI/CD, Google Test
Offensive Security
- Threat Modelling (MITRE ATT&CK, OWASP Top 10)
- Network Intrusion Detection Systems
- Wireshark, Nmap, Burp Suite
- NLP-Powered Phishing Simulation
- NIST Framework & Compliance
Skills Matrix
Active Operations
Education
Bachelor of Cyber Security
2024–2026Macquarie University · WAM 76.27
Digital Forensics · Network Security · Systems Security · NLP & Machine Learning · Privacy-Preserving Data Analysis · Cloud Computing
Diploma of Information Technology
2023–2024Macquarie University · WAM 71.75
Foundations in systems, networking, and software engineering.