AI agent security, explained simply
Every AI wants to be an Astra.
We give it a Leash.
AI agents can now read your files, run commands and spend money. We explain what they can really do, what goes wrong, and how to keep them on a leash.
Latest story · 26 Oct 2026
Black October: AI found a 10-year-old hole in the web
Two serious flaws sat inside PHP for more than ten years. This summer, AI helped two security teams find them. What happened, what the AI really did, and what to do today.
The framework
The LEASH
Five plain rules for keeping AI agents useful and under human control.
Limit
Give the agent only the access, time and budget the job needs.
Enforce
Turn rules into hard blocks the agent cannot quietly ignore.
Audit
Keep a clear record of what the agent saw, decided and changed.
Secure
Protect the agent, its tools and its data from tricks and misuse.
Human in the loop
A person decides wherever judgment and responsibility matter.
Interactive · drag to turn it
Your agent already has the keys.
Files, terminal, email, payments, browser, cloud. Every line is a permission. Switch on the LEASH and watch the blast radius shrink.
Audit log
Coming soon · private preview
WWW: your own AI mission control
A free kit that runs on your own computer: a notebook for every project, a night worker that writes up the day, a secret check that never lets a key slip into your notes, and a live dashboard. No tracking.
How we work
Every fact is checked against its primary source before a video goes out. Our host appears as an AI avatar with an AI voice, and we say so in every video. The story, the sources and the checking are real.