This isn’t a story about a brand-new kind of bug. The attackers found ordinary weaknesses (exposed credentials, weak passwords, trusted app connections) and let an AI agent chain them together faster than people could respond. The FBI says some intrusions went from first targeting to privileged access in under three hours.
Local Wiki Part 5 · The wiki was running. Why couldn’t I open the Learning Room?
On November 12, I tried to open the Learning Room, the interface I actually used to ask my local knowledge wiki a question. It was down. We had already tested the owner workspace behind it. That gave us confidence in a path I wasn’t using.
Building AI-SGAP: Inside the Engineering Work Since Our Last Update
Runtime validation, database integration, model evaluation, and the operational evidence still needed before partner release.
Building a Local Knowledge Wiki Part 4 Exact Evidence Partial Answers and Human Review
After we repaired retrieval, the model still failed a combined count-and-timeline question. It had relevant evidence, yet the application could not return a useful validated answer.
The next failure was surprisingly small: the model removed Markdown bold markers while copying source text.
Building a Local Knowledge Wiki Part 3 Source Imports and Retrieval Failures
The local model finally answered a real question correctly: how many domains are in the AI-SGAP framework?
The next question exposed a different problem. Asked how the risk-floor logic was calculated, the application returned “Not enough evidence.”
Building a Local Knowledge Wiki Part 2 Runtime Isolation and Graph Validation
In Part 1, our first MLX and Q4 attempts stopped under memory pressure. This installment covers the configuration and validation boundaries we used to make the next tests interpretable.
Building a Local Knowledge Wiki Part 1 Architecture and the First Memory Failures
This series follows the build as it develops. The system is still a pilot; several harder questions remain unresolved.
Selling AI Security Like Life Insurance
Selling AI security assurance is selling life insurance. You ask a buyer to pay today for protection against a day they would rather not picture. And the deal stalls for the same reasons.
Every insurance agent knows the pattern. The product makes sense, the prospect nods, and then the subject changes. Nobody enjoys rehearsing their own funeral.
What We Mean by “Harness” in AI
If you’ve spent any time around AI agents lately, you’ve heard someone say “it’s not the model, it’s the harness.” Release notes mention harness improvements. Benchmarks report scores “with our harness.” Engineers argue about whose harness is better.
It’s one of those words everyone uses and almost nobody stops to define. So let’s define it.
Same Container, Different Findings: What Trivy and Grype Taught Us
Trivy reported zero vulnerabilities. Grype reported ten findings at the same runtime-image checkpoint. Here is what that discrepancy taught us about package identification, advisory sources, filtering, and defensible container-security decisions.