Podcast
Honest lab notes on AI security, local deployment, and enterprise reality. No hype.
18 episodes published
Three recipes for a voice assistant that doesn't have to phone home. One takes ten minutes. One takes thirty. One takes a weekend and replaces Alexa entirely. I built one. I read the receipts on two more. Here's what each actually costs in time, money, and trade-offs.
EP.15 covered the source code. EP.16 covered the court fight. This is what people actually remembered.
Anthropic fought the Pentagon in court to protect its architecture. Days later, a missing .npmignore line published the entire blueprint to NPM. The fortress fell to a config file.
A missing .npmignore line exposed 512,000 lines of Claude Code source. An intern found it. Then axios got hijacked. One config file, two incidents, zero excuses.
Six products, a Pentagon lawsuit, and a self-graded safety plan. Everything Anthropic shipped in March and what it means for operators.
A CMS misconfiguration exposed roughly 3,000 unpublished assets. One described a model nobody was supposed to see yet. That was enough.
30 CVEs in 60 days. 38 percent of scanned servers with zero authentication. The protocol works. The guardrails don't exist yet.
Everything Anthropic shipped in March 2026. Sorted by who it actually affects.
I wired three MCP servers built by strangers into Claude Code. One sentence. Three external systems. No audit trail.
MCP 101 was the spec. MCP 102 is the behavior. What actually happens when you flip the switch in a real project stack.
Everyone is talking about MCP. Almost nobody has read the actual spec. I did. Here's what it says and what it doesn't.
Goldman Sachs automated back-office work with Claude. Your leadership team saw the headline. Here's what actually happened and what comes next.
Five steps to stop AI tools from leaking your information. A practical method for anyone using AI with sensitive documents.
Local doesn't mean safe. Here's the Lethal Trifecta that makes local AI agents a security problem nobody is talking about.
AI agents are making decisions in your name. When something goes wrong, the liability chain is longer than anyone admits.
What if your game remembered everything? A deep dive into persistent karma systems and what they tell us about AI memory.
I waited 14 days to do a 20-minute task because I was overthinking the setup. Here's what I learned.
The Nuclear Option. Why I stopped using cloud AI for sensitive work and built a local sovereign AI server on gaming hardware.
Subscribe on your favorite platform and get honest AI takes delivered every week.
Subscribe on Substack