AI × Security
Field notes and studies from the edge where AI agents meet security work: what I tested, what held, and what I'd check next.
3 pieces
AI × Security
The Model That Says Yes
I ran a refusal-reduced model on a spare laptop. It held more than I expected, reasoned less than I hoped.
AI × Security
Capital One’s VulnHunter: From Source Code to Live Attack
One AI security tool, one controlled target. What the tool carried, what I had to supply, and what a person still had to check.
AI × Security
Declared, Not Wired
A safeguard can exist in the design and still be missing from the route that acts. One test made that gap visible.