The AI Systems Security Specialist — Series Overview
A nine-part weekly series on securing AI systems — from tokens and context windows to trust boundaries, agent tool abuse, red teaming, governance, and the AI-assisted SOC.
Oussama Sahnoun
AI Agents & Tool Abuse — When the Blast Radius Moves at Machine Speed
Part 4 of the AI Systems Security Specialist series: agent autonomy as a risk multiplier — the confused deputy at machine speed, MCP as an attack surface, tool poisoning and rug pulls, memory poisoning, multi-agent trust, and the real 2026 agent incidents.
Secure AI Systems Engineering — Building So That Being Fooled Doesn't Matter
Part 3 of the AI Systems Security Specialist series: the 2026 state of the art in defending AI systems — information-flow control, dual-LLM patterns, spotlighting, deterministic egress control, capability scoping, and the honest answer to whether prompt injection can be solved.
Prompt Injection & Abuse Techniques — Attacking the Architecture
Part 2 of the AI Systems Security Specialist series: direct and indirect prompt injection, tokenizer-level filter evasion, multi-modal payloads, system prompt extraction, and the exfiltration channels that turn a text bug into a data breach — built around real 2025–2026 incidents including EchoLeak.
AI/LLM Systems & Security Architecture — The Foundation Every AI Security Assessment Starts From
Part 1 of the AI Systems Security Specialist series: the complete architecture of an LLM application — tokens, context windows, inference, the five-layer stack, orchestrators, RAG, embeddings, trust boundaries and sensitive data flows — plus a full worked security review of a vulnerable RAG chatbot.
Stormous Ransomware: A Complete Deep-Dive — Origins, Alliances, GhostLocker 2.0 Binary Analysis, and Full MITRE ATT&CK Mapping
TLP:AMBER — This analysis is prepared for the defensive benefit of the security community. All data is sourced from open-source intelligence: dark-web monitoring, public threat-intelligence tracker...
Cicada3301 Ransomware (Repellent Scorpius) — Complete CTI Report: Rust-Based RaaS Targeting ESXi, CVE Arsenal, IOCs & Full Defensive Playbook
Comprehensive Cyber Threat Intelligence report on Cicada3301 (Repellent Scorpius) — a Rust-based RaaS with ALPHV/BlackCat code overlap targeting Windows, Linux/ESXi, NAS, and PowerPC. Full MITRE ATT&CK TTP mapping, CVE arsenal, confirmed IOCs, YARA & Sigma rules, and a complete defensive playbook.
Cyber Threat Intelligence: The Art of Thinking Like the Enemy
A deep technical and strategic guide to Cyber Threat Intelligence — from the lifecycle and MITRE ATT&CK heatmap coverage to IoC enrichment pipelines, cognitive bias in triage, and structured case study analysis of SolarWinds and NotPetya.
I Spoke at Black Hat MEA 2025 in Riyadh — Here's Everything I Know About AI and Threat Intelligence
The full technical writeup of my Black Hat MEA 2025 session in Riyadh: how LLMs and RAG architecture are transforming threat intelligence, a deep dive into ThreatLens (open-source), and the skills that define the next generation of security engineers.
Web LLM Attacks — Part 2: Advanced Attack Surfaces in AI Systems
A deep dive into advanced attack surfaces in modern AI systems: AI agents, RAG pipelines, embeddings, MCP tool surfaces, supply chain attacks, and AI infrastructure exploits.