A Robot Is Just a Phone with Legs

Originally published in The Access Control Collection (TACC) on September 25th, 2026.

After my last piece, an engineering friend texted me to point out a logic gap.

He wrote: “Fine, I root the phone. The access token is still sitting on it. I present it. The door opens. What exactly did you win?”

He was right.

I skipped the technical plumbing to make the article readable. (My mistake. Let’s suffer through the plumbing).

The question “what breaks if I root the phone?” is really shorthand for a much scarier question:

Can the holder now open a door they weren’t already allowed to open?

If the lock is online, it’s easy. The server issues a token for this exact door, for this exact second, for one use. The lock just refuses anything it has already seen.

A lock doesn’t need a clock. It needs memory and a calendar that only turns forward.

But take the network away, and things get ugly.

The lock can no longer tell a fresh token from a hoarded one. A token issued this morning and a token lifted last month look identical.

So the lock must never accept a stored token on its own. Whatever carries the token has to bring something the lock can check at the door: something you know, something you have, something the lock itself prepared. There are several ways to do this. I’ve built one.

The IT world learned this the hard way with bearer tokens. They fixed it with “proof of possession”—binding the token to something the thief doesn’t have. NIST just finished writing guidance on it.

Physical security folks hate this comparison.

When I bring it up, they tell me, “It’s different! We have cameras. We have guards. A teenager in Iran can’t hack a physical door.”

Which is a comforting bedtime story, right up until you realize that if you protect something actually valuable, people will show up to take it.

We need to learn IT’s lesson now, because the “holder” of the token is no longer human.

Michael Stuer wrote in last week’s issue a brilliant piece about the automated agent that manages access. It reads a work order and extends a contractor’s access at 2 AM. No human in the loop.

NIST says the exact same token rules apply to AI agents, and freely admits it doesn’t yet know what else agents will need.

Neither does anyone else.

Or look at the robots walking up to readers at GE facilities.

A robot is just a phone with legs. Same trust question, no human behind it.

You cannot interrogate an AI agent about why it issued a token. You cannot ask a robot if it’s lost.

Soon, there is going to be a lot of intelligence in the access chain. Some of it will be wrong at 2 AM.

The lock is the one place in that chain where nothing has to think.

It has a key that never left, a memory that only moves forward, and a single rule: prove it, or the door stays shut.

Let everything else be clever.