Blog

Jul 14, 2026

Bleeding Llama (CVE-2026-7482): Your Local LLM Is an Unauthenticated Memory Leak

A heap out-of-bounds read in Ollama’s GGUF loader lets an unauthenticated attacker exfiltrate process memory — prompts, secrets, env vars — in three API calls. ~300,000 servers listen on 0.0.0.0 by default. Everyone tells you to patch. Nobody tells you how to detect it. This does.

IP Reputation Is a Fragile Detection Signal — What I Learned Building a 4G Residential Proxy

Most WAFs lean hard on IP reputation to tell humans from bots. Building the evasion side — an old Android phone as a 4G residential proxy — showed me exactly where that signal collapses, and why rate and sequence behavior outlast any IP allow/deny list. A detection-engineering post-mortem, with the full stack included.

Your 'Impossible' Bypass Has an Expiry Date

A geo-fence I’d documented as ‘impossible to bypass — case closed’ quietly became trivial, because one variable I treated as constant wasn’t. This is a field note on reverse-engineering a marketplace’s geo-filter in real time, the three undocumented gotchas that decide whether it honors your coordinates or silently ignores them, and the bigger lesson: every bypass you write down has a shelf life, because the other side’s infra moves while your notes sit still.

Your Health Check Is Lying to You

Availability and correctness are two different questions, and most monitoring only answers the first one. I walk through how a system can pass every health check while serving confidently wrong data, why ‘green’ is the most dangerous color on a dashboard, and the one query that turns a liar into a witness.

The Hidden Cost of Agent Loops — A Security and FinOps Take

Loop engineering is the title of the week. But the engineers promoting it work at the token dispensaries and never see the bill. I break down the loop hype from two angles the hype skips — what it costs, and what it does to your control model — and land on the rule that actually matters: the agent executes decisions, it does not get to make them.

26% of AI Agent Skills Are Vulnerable. I Audited My Own.

A study of 42,447 agent skills found 26.1% vulnerable and 5.2% likely malicious. I ran my own 29 skills against the SkillSpector taxonomy. They came back clean — but the audit exposed three real gaps in my defense model, and the fix became a new tool.

AI Can Find 10,000 Vulnerabilities in Weeks. Who's Going to Patch Them?

When AI finds vulnerabilities faster than humans can patch them, the limiting factor stops being detection and becomes verification and remediation. Amdahl’s law applied to security operations: a security engineer’s read on Anthropic’s recursive self-improvement essay.

The IC Score: An AI-Powered Formula for SOC Alert Triage

A practical breakdown of the IC Score formula: 7 weighted signals, asset criticality multipliers, and decision thresholds that reduced our mean triage time from 15 minutes to under 2 minutes.

PewDiePie's Odysseus: The Security Take on Running an Agent That Reads Your Email and Executes Code

Odysseus combines command execution, email reading, and web access in a single local agent. It’s a textbook case of the ’lethal trifecta’. ‘It’s local’ protects your privacy, but not from prompt injection. The security take and how to install it right.

Designing Guardrails for AI Security Agents in Production

AI agents with access to your SIEM, identity provider, and containment actions are privileged processes. Here is how I design guardrails to keep them safe in production.