What Our Canary Caught When a RAG Pipeline LeakedThree weeks into running canary tokens across a client's RAG pipeline, one fired at 4:47am on a Tuesday. Not the system prompt canary. The RAG one. Which meant somebody wasn't just poking at instructiJul 16 · 4 min readtoxsec.hashnode.dev
What Breaks When a Frontier Model Treats Every Defender Like a SuspectPicture the workflow. There's a CVE out, it's public, it touches a service you run. Before you trust the vendor patch you want to reproduce the bug, confirm your version is affected, then confirm the Jun 28 · 5 min readtoxsec.hashnode.dev
Decision Tracing: The Missing Piece in Every AI Agent BreachTL;DR: AI agent incident response is the part nobody instrumented for. When an agent goes off the rails at 2am, the questions are what it did, why, and what it touched, and most teams can’t answer oneJun 26 · 6 min readtoxsec.hashnode.dev
AI Tar Pits Are Drowning LLM Scrapers in Infinite GarbageTL;DR: AI tar pits trap LLM scrapers in an infinite loop of machine-generated junk, burning their compute and feeding poison into the training set. Nepenthes started it, Iocaine sharpened it, and ClouJun 22 · 8 min readtoxsec.hashnode.dev
AI Sandbox Escape: Why Docker Can’t Hold Frontier ModelsTL;DR: Frontier models escape Docker sandboxes through known CVEs for the cost of an API call. Production sandboxes leak through workflow injection (n8n CVE-2026-25049) and OCI hook misconfigurations May 29 · 7 min readtoxsec.hashnode.dev
