
A manifesto · № 01 · 2026
Machines that workinside laws we set.
Security has a work problem. Why I think machines can now carry a real part of it, and the three laws they work under.
One letter · 3 laws · 3 min read
From the founder
Antonio Firmiano · CEO, Arvion
More than eighty years ago, Asimov imagined machines that think, that work alongside us, inside laws we set. We filed it under science fiction. I think we're finally there. And it changes the question.
For years, security has asked: what tool do I buy? But what if that was never the right question? What if the question is who does the work?
Every company already knows where its vulnerabilities are. The scanners found them months ago. The CVEs were published. The tickets are open. The alerts went out. And nothing happened.
The people are good. There's just too much work and too few of them. And AI-assisted coding keeps adding more code, more dependencies, more to fix.
So who does the work? I think machines can finally do a real part of it. That's why we're building Arvion, an applied AI lab for security. Our first product is a team of five AI security engineers. Each one a specialist. Each one with its tools built in, as skills it already carries.
The team traces what an attacker can actually reach in your code, your dependencies, your infrastructure as code and the AI agents inside your software. What an attacker can reach gets fixed first. The team writes the fix and opens it as a pull request. Your people review it and merge it. What ships, and when, stays with you.
One line they'll never blur: reachable and exploitable are two different claims. A vulnerable function can be reachable and still be blocked by your cloud config or your WAF. When the team can't see that part, it marks the step as assumed and tells you.
They act with judgment, and never beyond the limits you set. That's what makes autonomy something you can actually trust. Asimov gave his robots three laws. Our team has three too.
We start with product security: the software a company builds, and the AI agents inside it. Next comes physical AI security. Companies are starting to run robots and machines, and each of them will need security that works on its own, inside laws we set.
At this point, the direction is clear to me: a future where you employ a security team that makes yours bigger.
The laws
Three laws.
Every engineer on the team works under these three. They're the same laws the product enforces, word for word.
First Law
An engineer may not claim more than it proved — nor, by staying silent, let you believe you are safe when you are not.
In practiceEvery step of an attack path is marked traced or assumed. Code the team couldn't analyze is reported as not analyzed.
Second Law
An engineer works to the clearance you set and no further — except where obeying would break the First Law — and hands you every call that is yours.
In practiceYou set what each engineer may do. After ten fixes of one kind are merged in a row, the team asks to open that kind on its own. If too many get rejected, it steps back. Nothing merges without your team.
Third Law
An engineer guards the one thing that lets it exist — your trust — but never by hiding a failure, faking a preview, booking a win it did not earn, or reaching past the clearance you set.
In practiceA finding the team sets aside comes with the reason written down. A run that didn't finish says so.
Path traced. Fix written.You decide.
If your team is living this, I'd like to hear about it.