Authority on Paul Mozaffari

The Reversibility Test: Grant AI Autonomy by Undo, Not by IQ

Fri, 26 Jun 2026 00:00:00 +0000

We decide how much freedom to give an AI system by asking how good it is.

How accurate is the model. How well did it demo. How impressive was the benchmark. Then, satisfied it’s smart enough, we wire it into the systems that matter and let it act.

That’s the wrong axis. Accuracy tells you how often the system is right. It tells you nothing about the cost of the day it’s wrong — and in production, everything eventually has that day.

I spent 28 years in network security. The lesson that outlasted every technology I touched is this: you don’t get to choose whether things fail. You only get to choose how far the failure travels and whether you can walk it back. The first half of that is blast radius. This is the second half.

The question is not “is it smart?” It’s “can I undo it?”

Here’s the reframe I give the leaders I work with. Before you let an AI system take any action on its own, ask two things about that specific action:

Can we undo it — and how fast?

That’s the Reversibility Test. Two questions, applied per action, not per system. And it inverts how most teams hand out autonomy.

Because a dumb action that’s instantly reversible is safe to automate all day long. A brilliant action that can’t be undone is exactly the one that needs a human in front of it. Intelligence is not the thing that should earn an agent the right to act alone. Reversibility is.

We knew this before AI. It’s why databases have transactions you can roll back. It’s why every competent change request ships with a rollback plan before it ships the change. It’s why the irreversible commands — wipe the array, push to prod, fire the missile — get the two-person rule. We never granted authority on the basis of how confident the operator felt. We granted it on the basis of what happened if they were wrong.

Agents need the same discipline, and almost nobody is applying it.

The Reversibility Ladder

“Can we undo it” isn’t a yes or no. It’s a ladder. Every action an agent can take sits on one of four rungs, and the rung — not the model’s accuracy — should decide how much autonomy it gets.

Tier 0 — Reversible. Undo is instant and free. Drafting a reply, suggesting a tag, proposing a change. If it’s wrong, you delete it and move on. Let the agent run autonomously. This is where the productivity actually lives, and most teams under-automate it because they’re scared of the tiers above.

Tier 1 — Recoverable. Undo exists but it costs time or effort. A config change with a rollback path, a database write you have a backup for. Allow it autonomously — but only if the rollback is built, tested, and fast. An undo you’ve never rehearsed is not an undo. It’s a hope.

Tier 2 — Compensable. You can’t undo it, but you can offset it. You can’t un-charge a card, but you can refund it. You can’t un-send a wrong answer, but you can issue a correction. Allow with a compensating control and a human notified — someone has to know the offset is needed, or it never happens.

Tier 3 — Irreversible. No undo, no offset. Money wired to an external account. Data deleted with no backup. An email sent to a customer. A public statement posted. A production resource destroyed. A human approves, every time, no exceptions. This is the rung where “the model is usually right” stops being a defense and starts being the epitaph.

The work is simple to describe and uncomfortable to do: take every action your agent can take, and put each one on a rung. The discomfort is the point. Most teams have never made that list. They deployed the capability and assumed the accuracy would hold.

Why this pairs with blast radius

If you’ve seen my Blast Radius Test, reversibility is one of its four questions — Reach, Authority, Reversibility, Detection. I’m pulling it out and going deep on it here for a reason: it’s the most actionable of the four. You rarely get to shrink an agent’s reach without gutting its usefulness. But you can almost always gate it by reversibility without touching what it’s good at.

Run them together and you get the grid that actually matters:

Blast radius asks how far does the damage spread. Reversibility asks can I pull it back. An action that’s wide-reaching and irreversible is the one that should never run without a human — and it’s the one teams wave through because the demo was clean. An action that’s narrow and reversible is free to automate aggressively. Most governance effort is spent in the wrong corners of that grid.

What this changes for the person signing off

If you’re accountable for an AI deployment, you don’t need to understand the model’s architecture to govern it. You need one artifact: the list of actions the agent can take, each one assigned a reversibility tier, with Tier 3 explicitly gated behind a human.

If your team can’t produce that list, that is the finding. It means autonomy is being granted by vibe — by how good the thing seems — instead of by what it costs when it’s wrong.

So before your next agent goes live, the question isn’t “how accurate is it.” It’s: for everything this agent can do on its own — can we undo it, and how fast?

Stop granting autonomy by IQ. Grant it by undo. That’s the version that survives operational reality.

Part of the AI Security collection. Related: AI Safety Has Never Worked a Change Window · The Zero-Trust Agent.

Putting AI into production and want production-scarred eyes on it? I run private AI-security briefings for leadership teams — message me on LinkedIn and mention “briefing.”

Advice to My Younger Self (If I Knew What I Know Now)

Fri, 05 Jun 2026 00:00:00 +0000

You thought being smart meant understanding complexity. You spent years proving you could hold more of it than anyone else in the room.

You’ll learn the opposite. Being smart is taking complexity out without losing the truth underneath it.

And stop collecting tools. You’ve got drawers full of them. Spend the time building systems instead. That’s the thing you’re actually good at.

1. Your edge isn’t what you studied. It’s what you joined together.

You are not “just” a network engineer. Don’t let anyone, including you, file you under that.

You design systems under real constraints. You translate between the business and the infrastructure. You find the clear line through environments that are honestly a mess. And you keep going long after the novelty wears off, which is the part most people quit.

That combination is rare. Authority lives at the intersections, not on the title.

2. You’ll undervalue your experience, because from the inside it just feels like Tuesday.

Twenty years in infrastructure feels routine to you. Of course it does. You’ve been standing in it the whole time.

From the outside it’s something else entirely. It’s pattern recognition at a scale most people never reach.

Most people learned the technologies. You learned the failure modes. That’s not the same job. That’s architectural thinking.

3. You’ll try to get smarter when what you actually need is leverage.

Reading more will help, a little. But publishing helps more. Teaching helps more than that. Building frameworks other people can use helps most of all.

Knowledge sitting in your head doesn’t compound. It only compounds once it’s out of you and in the world.

4. You’ll think the work is about networking and security. It isn’t.

It’s about making decisions when you can’t see the whole board.

Security taught you something most professions never teach: assume you’re already breached, design anyway, and keep the thing running while you do. Sit with how unusual that is.

That mindset carries straight into AI, consulting, leadership. It travels further than you expect.

5. You’re not late. That’s a story you tell yourself.

You’re not late to AI. You’re early to the part that actually matters — applied governance and what happens when these systems meet reality.

The young engineers understand the models. Almost none of them understand the systems around the models. You understand both. That’s the whole game.

6. Complexity will tempt you, because it looks like competence.

Don’t fall for it. The people you respect most do the opposite. They simplify the architecture. They simplify the explanation. They simplify the decision. They simplify what gets your attention at all.

And be clear with yourself about what simple means. It isn’t leaving things out. It’s compression.

7. Productivity was never about doing more.

It’s about lining the work up with who you actually are.

That itch to build systems — time alignment, learning pipelines, accountability dashboards, ways to organize what you know — that’s not a hobby you should feel guilty about. That’s your signature. Lean into it.

8. Teaching isn’t a side thing you’ll get to later.

It’s the multiplier. It’s the whole point.

When you explain a system clearly, executives start trusting you, engineers start following you, clients start hiring you, and an audience quietly builds around you. Your documentation becomes your reputation. Write more of it down.

9. Your authority comes from synthesis, not specialization.

Specialists solve the problem in front of them. Synthesists decide which problems are even worth solving. You’re the second kind. Stop apologizing for it.

You’re moving from engineer to architect to advisor to framework creator. Don’t stall that. Don’t let comfort stall it. Push it faster.

10. Your kids aren’t competing with your ambition.

They’re shaping it. They’re the reason it has a direction worth having.

A lot of people try to become extraordinary before they’re willing to be present. You’ll be tempted to do it in that order too. Reverse it.

Presence gives you clarity. Clarity gives you leverage. Leverage gives you freedom. That’s the actual sequence.

11. You don’t need anyone’s permission to build intellectual property.

A framework counts. A methodology counts. An audit model counts. A checklist counts. A course counts.

Publishing is just the act of turning experience into something you own. Start sooner.

12. When something feels like overthinking, it usually means ship.

Your instinct is analysis. It always has been. But your edge is going to be cadence, not depth-of-deliberation.

Write before you’re ready. Publish before it’s perfect. Teach before you feel qualified. Confidence isn’t a personality trait you’re missing. It’s just evidence, stacked up, of finishing hard things. You build it by finishing.

13. The pivot you keep waiting for is already underway.

You’re moving from technical contributor to system designer to knowledge architect. It’s happening right now, quietly, whether you’ve named it or not.

Most people miss the moment their own transition begins. You didn’t. That awareness is leverage all by itself.

14. Stop mistaking preparation for progress.

If no one outside your notebook can see the work, it doesn’t exist yet. Sit with that one.

Make things that get seen. Things that get reviewed, rejected, improved, reused. That loop is where the real learning is.

Visibility compounds faster than perfection ever will. So ship it early, ship it in public, and ship it again next week.

15. Don’t try to out-run AI on speed.

You’ll lose. It already won that race. Don’t enter it.

Compete where you’re still scarce: judgment, knowing what can go wrong, holding the context, owning the outcome when it’s your name on it. Your value was never how much you could produce. It’s how reliable your decisions are.

16. Become the calmest, safest person in the room.

Every era has one technical thing that frightens the people in charge — the thing they’re accountable for but can’t quite see. Right now it’s AI.

If you can say plainly what it does, what it breaks, what it puts at risk, and what it makes possible, you stop being a vendor and start being infrastructure. And infrastructure outlives the hype cycle every time.

17. Treat your time like a security boundary.

You get 168 hours a week. That’s it. If you don’t spend them on purpose, someone else will spend them for you.

Defend the things that don’t defend themselves: your health, being there for your family, the windows where you do deep work, the time you set aside to learn. Attention is the credential underneath every other credential you’ve earned.

18. Use the leverage tools early, not once they’re respectable.

The people who refuse new tools to prove they’re doing it “the proper way” fall behind. Quietly, and then all at once.

Pick up the exoskeleton. Let the machines carry the load on your thinking, so your thinking can go to where the decisions are.

19. Your reputation is just your execution history, stored.

Confidence isn’t who you are. It’s what you remember. Specifically, it’s the memory of finishing difficult things, over and over, until finishing feels normal.

Stack those. Almost everything else you want follows from that one pile.

When you step back and look at the whole shape of it, here’s what it says about you.

You treat your life as one system, not a pile of competing priorities. Health, family, career, learning, execution — you keep trying to make them fit together instead of trading one off against another. That’s rarer than you give yourself credit for.

And you optimize for alignment, not intensity. Most people chase effort and burn out proving it. You’re building coherence that holds across years. That’s what gives you staying power when other people run out of road.

So here’s the one move that matters most right now: lean all the way into translating AI risk for the people who have to lead through it. Not tooling tutorials. Not prompt libraries. Not productivity tricks. Decision-layer clarity, in language a leader can act on.

That’s where your 28 years stop being a backstory and start being an unfair advantage. Go use it.