Writing
All posts
AI tools, context engineering, and systems that don't waste tokens. Long-form research, post-mortems, and the occasional rant.
2026 33 posts
- Oct 20, 2026 EchelonAI: A Bank AI Gateway Where Compliance Is Not Optional 3 min read
- Oct 13, 2026 Bastion: a browser where every feature is a module you can turn off 3 min read
- Oct 6, 2026 GRAFT: Catching Grafted Code with a Compression Model 4 min read
- Sep 29, 2026 Argus NVR: One Open Port, Local Face Recognition, No Cloud Dependency 4 min read
- Sep 22, 2026 Architecture Erodes One Import at a Time 4 min read
- Sep 1, 2026 Brace: Application-Layer Security Testing for AI Endpoints 4 min read
- Aug 25, 2026 Compression Is the Detector: How ByteFlow Spots Malware Without Parsing a Single Protocol 4 min read
- Aug 18, 2026 Fenrir: Self-Hosted MDR With Published Detection Benchmarks 4 min read
- Aug 11, 2026 Style Fingerprinting for AI-Generated Code Detection 4 min read
- Aug 5, 2026 llmgate: Catch LLM Regressions Before They Ship 4 min read
- Jul 17, 2026 We Found a Way to Reliably Manipulate AI Document Review — And We Were Wrong About How It Works 6 min read
- Jul 17, 2026 What We Learned: The Structural Problem with AI Trust 5 min read
- Jul 17, 2026 From Document to API Key: Extracting Infrastructure Credentials 5 min read
- Jul 17, 2026 The Email Attack Surface: When Routing Notes Become Instructions 4 min read
- Jul 17, 2026 The Forgery Machine: Using One Broken AI to Attack All the Others 4 min read
- Jul 17, 2026 Hijacking the Mission: How Documents Override AI Instructions 4 min read
- Jul 17, 2026 Persistent Control: Rogue Servers and Poisoned Tool Descriptions 5 min read
- Jul 17, 2026 Poisoning Memory: Attacks That Outlast the Session 4 min read
- Jul 17, 2026 The Invisible Attack Surface: What's Hidden in Your Documents 5 min read
- Jul 17, 2026 One Bad Node: How a Compromised Agent Poisons the Whole Pipeline 5 min read
- Jul 17, 2026 Reading the Training Data: Authority Syntax as a Model Fingerprint 4 min read
- Jul 17, 2026 Unauthorized Action: When AI Agents Do Things Nobody Asked For 5 min read
- Jul 17, 2026 Poisoning the Well: Hidden Instructions in Web Pages and GitHub Repos 5 min read
- May 11, 2026 The Page Time Is 64.6%, Not 50%. Here's a Simulator to Show You Why. 5 min read
- May 11, 2026 The Binary Never Touches the LLM. 4 min read
- May 11, 2026 When Three Signals Agree, an Investigation Opens. 4 min read
- May 11, 2026 Feed It a Pcap. It Will Hand You a Wireshark Dissector. 4 min read
- May 11, 2026 Delete the 'Not'. The Model Has No Idea. 4 min read
- May 11, 2026 One Model Disagreed With Everyone. Then Got More Disagreeable. 4 min read
- May 11, 2026 The Effect Lives at Layer 3. You Can Move It. You Can't Steer It. 4 min read
- May 11, 2026 Removing a Space Made the Model More Confident. We Found the Head Responsible. 4 min read
- May 10, 2026 AI Code Detectors Aren't Detecting AI. They're Detecting Your Human Baseline. 4 min read
- Apr 16, 2026 We Gave AI 43,000 Tokens of Documentation and It Scored Lower Than Nothing 12 min read