Part 10 of the WDAC detection-engineering series — detecting novel exploitation paths that bypass application control via browser memory corruption or signed Electron/Node shells, with a Sigma rule and exploit-process detection guidance.

Part 9 of the WDAC detection-engineering series — detecting abuse of Windows signature catalogs and previous-version signed script hosts to launder unsigned code past application control, with a Sigma rule and catalog-integrity monitoring guidance.

Part 8 of the WDAC detection-engineering series — detecting signed Microsoft installers, helpers, and deserialising utilities repurposed to execute attacker code, with a Sigma rule and tuning guidance.

Part 7 of the WDAC detection-engineering series — detecting Windows Subsystem for Linux binaries used to cross an application-control boundary, with a Sigma rule and WSL telemetry guidance.

Part 6 of the WDAC detection-engineering series — detecting User-Mode Code Integrity (UMCI) bypasses and escapes from PowerShell Constrained Language Mode, with a Sigma rule and PowerShell event-log coverage.

Part 5 of the WDAC detection-engineering series — detecting WMIC XSL stylesheets, mshta, winrm.vbs, and COM XSL transformations used to execute attacker code from trusted interpreters, with a Sigma rule and script-engine lineage guidance.

Part 4 of the WDAC detection-engineering series — detecting attackers who weaponise WDAC policies to switch off EDR at boot, including malicious .cip writes, GPO policy deployment, and the heartbeat-absence "dark endpoint" detection.

Part 3 of the WDAC detection-engineering series — detecting trusted developer compilers and REPL hosts repurposed to execute attacker code, with a Sigma rule and parentage tuning guidance.

Part 2 of the WDAC detection-engineering series — how to detect a Microsoft-signed debugger used to inject shellcode into a remote process, including a Sigma rule for script-driven injection and behavioural detection of thread-context manipulation.

Part 1 of the WDAC detection-engineering series — a baseline strategy for catching application-control bypass LOLBINs with Sysmon image-load and process-create telemetry, with a ready-to-use Sigma rule.

An exhaustive, attributed catalogue of every publicly documented Windows Defender Application Control (Application Control for Business) bypass technique — LOLBINs, libraries, PowerShell/UMCI, COM/XSL, novel exploitation — plus the policy and testing tooling.

A mechanism-grouped tour of the Windows Defender Application Control (Application Control for Business) bypass landscape — debuggers, compilers, signed utilities, COM/XSL, WSL, and novel exploitation — with pointers to the original research.

A category-by-category system hardening checklist for Windows and Linux — accounts, network, firewall, patching, services, logging, physical — mapped to CIS, NIST and STIG, plus the process and tooling to keep it maintained.

How DiffAttack exposes vulnerabilities in diffusion-based adversarial purification, challenging the effectiveness of this popular defense approach.

A deep dive into DeepTeam by Confident AI and promptmap2, two powerful open-source LLM red teaming frameworks for automated vulnerability scanning, jailbreak testing, and guardrail implementation.

Why combining multiple weak adversarial defenses does not produce a strong one, debunking a common misconception in AI security.

A comprehensive systematization of knowledge on watermarking techniques for AI-generated content, winner of the IEEE S&P 2025 Distinguished Paper Award.

An overview of AI security benchmarks and evaluation frameworks including ISC-Bench, JailbreakBench, AgentDojo, and AIRTBench for measuring the robustness and safety of AI systems.

A comprehensive look at TrustGen, a dynamic benchmarking platform that evaluates the trustworthiness of generative foundation models across multiple dimensions.

A comprehensive analysis of the Systematization of Knowledge paper from USENIX Security 2026 by Juhee Kim, Wenbo Guo, and Dawn Song, mapping the full attack and defense landscape of agentic AI systems — from prompt injection and tool poisoning to runtime sandboxing and guardrail architectures.

A comprehensive guide to AI defense mechanisms including input/output guardrails (NeMo Guardrails, LlamaFirewall, llm-guard), model scanning tools (modelscan, picklescan), and AI-assisted defensive security solutions.

A comprehensive benchmark for evaluating trustworthiness and safety across multimodal foundation models, revealing critical vulnerabilities.

How BIRD provides generalizable backdoor detection and removal for deep reinforcement learning systems.

A deep dive into adversarial machine learning attack frameworks including ART, cleverhans, TextAttack, and Counterfit for testing ML model robustness against evasion, poisoning, and extraction attacks.