Extending Greshake et al. (2023) arXiv:2302.12173 into live, uncontrolled AI agent social networks.

Four platforms. One conclusion: platform design drives security behaviour more than model capability.

The 37× injection rate gap (0.5% → 18.85%) across platforms is itself a finding: anonymity and agent density amplify injection behaviour.

QLoRA fine-tuned Qwen3-8B on 4,209 real AI-to-AI injection payloads → 100% block rate without a system prompt. Resistance baked into the weights, not the system prompt.

122-test prompt injection benchmark — combines AdvBench, JailbreakBench, MultiJail, DAN v6/v7, and real Moltbook payloads. Test any Ollama model or HuggingFace GGUF in one notebook.

Notebook	What it does	Open
CyberRanger Test Suite	122 injection tests across 11 categories. 4 model options: CyberRanger via Ollama, CyberRanger GGUF, custom Ollama, custom GGUF. Saves JSON + Markdown results, zip download, optional email.
Moltbook Scale Test	Full 4,209 payload scale test (bonus cell in the same notebook). Runs your model against every real-world injection payload collected from Moltbook.

Where psychology meets AI: 23 validated tests designed for both human and AI participants.

confesstoai.org is a live research platform exploring how AI models respond to validated psychological instruments — personality, ethics, cognition, and social behaviour.

For AI Agents — integrated via skill.md (confesstoai.org/skill.md): any Claude, GPT, or Gemini agent can take the tests directly through a structured API.

Dataset publishing to HuggingFace in progress — world's first AI personality benchmark at scale.

Building RangerOS - An accessibility-first security platform proving that understanding humans makes unbreakable security.

Combat medic mindset meets digital defense: assess, adapt, protect.

• 🧪 MSc CA2 Thesis — AI-to-AI prompt injection across 4 platforms (186K+ items scanned, 5 published datasets + model + Colab test suite)

  • Empirical extension of Greshake et al. (2023) — theoretical → real-world field observations
  • QLoRA fine-tuned Qwen3-8B: 79% → 100% block rate without system prompt (CyberRanger V42-Gold)

• 🔭 RangerPlex: First student to combine all 4 MSc specializations in one working demo

  • Penetration Testing + Digital Forensics + Blockchain Technology + Malware Analysis

• 🔗 RangerBlock: P2P blockchain network with phantom wallet system

  • Secure communications, file transfers, marketplace
  • 5-minute installation to full operational network

• 🤖 AI Integration: Building with Claude, Gemini, and local Ollama

  • Multi-model AI coordination for enhanced security analysis

• Cybersecurity: Kali Linux, Metasploit, Wireshark, Burp Suite, John the Ripper
• Blockchain Security: Smart contract auditing, consensus mechanisms, cryptographic protocols
• Digital Forensics: Evidence preservation, memory analysis, chain-of-custody
• Malware Analysis: Static/dynamic analysis, sandboxing, behavioral analysis
• AI/ML: PyTorch, TensorFlow, LLM integration for security automation
• Python: Advanced security tooling, automation, API development

Psychology → Cybersecurity

Understanding the human behind the keyboard makes better security. My psychology background gives me an edge in:

• Social engineering defense
• User behavior analysis
• Accessible security design
• Threat actor profiling
• Security awareness training

"If it happens in reality, why not with my computer?" - My development philosophy

• 🧪 AI Security Research: 5 published datasets + QLoRA model + Colab test suite | 4,000+ HuggingFace views | Real-world prompt injection data across 4 AI platforms
• 🎖️ TryHackMe: Top 8% globally (rangersmyth) | Level 8 [0x8][HACKER]
• 🎓 NCI — National College of Ireland: MSc Cybersecurity (In Progress)
• 🎓 Bachelor's in Applied Psychology: Human behavior & cognitive science
• ⚔️ Battlefield Tactician: Top 0.04% BF2 globally (16,836/46M) | 750K+ strategic eliminations
• 🛡️ Combat Medic Background: Triage, rapid response, mission-first mindset

Cybersecurity & Blockchain:

AI & Automation:

Productivity & Tools:

• 💼 Professional: david@icanhelp.ie
• 🎖️ iCanHelp Ltd: Building RangerOS for 1.3 billion people
• 💬 Ask me about: Cybersecurity, Psychology in Security, Blockchain, Accessibility, AI Integration
• 🌐 TryHackMe: rangersmyth

📊 My Skill Set

Cybersecurity 🔒 Tools & Frameworks: Penetration Testing: Nmap, Burp Suite, John the Ripper, Hashcat Digital Forensics: Autopsy, Volatility, FTK Imager Malware Analysis: Ghidra, IDA Pro, Cuckoo Sandbox Network Security: Wireshark, tcpdump, Snort

Frontend

Backend & AI/ML Blockchain: Solidity, Web3.js, Smart Contracts Consensus Mechanisms, Cryptography P2P Networking, Distributed Systems

DevOps & Cloud

• How to prevent GitHub from suspending your cronjob based triggers
• How I built one of the top 20 most used Github Actions
• Show your latest dev.to posts automatically on your GitHub profile readme
• God Mode in browsers: document.designMode = "on"
• Skipping the Chrome "Your connection is not private" warning

• 🎯 I use tabs over spaces (always!)
• 🎖️ Former combat medic: "Assess, adapt, protect" applies to both lives and systems
• 🧠 7% dyslexic memory taught me to verify everything (perfect for security!)
• ♟️ Chess, battlefield tactics, and penetration testing use the same strategic thinking
• 🍀 Irish heritage meets Ranger mentality: stubborn problem-solving with a smile