# RFID for the CISO

> What a stranger with a handheld reader can read, rewrite, copy or kill on a rack of tags, what each protection changes, and why the reader on your network is the bigger problem.

- **Page** - https://iotready.com/insights/rfid-for-the-ciso
- **Author** - Tej Pochiraju
- **Published** - 2026-11-06
- **Topics** - Technology, Security

Stand in an aisle with a handheld reader and point it at a rack.

The tags answer. They don't check who's asking, because a passive UHF tag has no way to know. Whatever your own dock door gets, the stranger gets.

Film: [A handheld reader lighting up a rack of tags, the two Gen2 passwords, what Impinj and NXP add, and the reader as the bigger surface](/media/rfid-13-rfid-for-the-ciso.mp4)

_A tag answers whoever asks. Then the passwords, the chip makers' extras, and the reader on your network._

I'm not going to tell you how anyone does anything here. What matters for a security review is what each choice stops and what it doesn't, and that fits in a table. There's a page that lets you fill it in: [What can a stranger with a handheld see?](https://iotready.com/tools/stranger-with-a-handheld)

## Nothing set

A Gen2 tag has two passwords. Each is 32 bits. The access password lets a reader write to locked memory. The kill password switches the tag off for good. In the GS1 standard both are zero until somebody programs them.

A tag with a zero kill password won't run a password kill. That sounds like protection. But with the passwords unlocked and the access password at zero, a reader can set a kill password first and then use it. The standard knows this. It recommends unique kill passwords, and it recommends against a zero access password.

So with nothing set, a stranger can read the EPC and the TID, rewrite the EPC, copy it to a blank tag and kill the tag. All four.

## Locked

Set a non-zero access password, lock the EPC bank and lock the passwords, and the writes stop. A locked password can't be read or written without the access password, unless it was permalocked. A locked EPC bank is unwriteable but still readable, so the stranger still gets the number.

That's the part people miss. Locking isn't hiding.

A 32-bit password is also a small space. The IEEE RFID 2017 keynote I'm leaning on says so. The GS1 standard's own advice is unique kill passwords, tag by tag, not one value across the warehouse.

## Silent

Impinj's Protected Mode is the first thing here that hides the number. With it on, the tag doesn't respond unless the reader sends the correct password. Your own readers send that password over the air, so someone eavesdropping close to them isn't covered. To a reader without it the rack looks empty. Impinj lists it for its M700 and M800 series chips, and says its readers turn it on and off with ordinary Select, Access and Write commands.

Your own readers hold the password and read as normal. The stranger gets nothing to read, rewrite, copy or kill, because the tag never answers him. The same small password space applies, and the page marks it.

## Authenticated

NXP's UCODE DNA is a different idea. It reads like any other tag, and it can also answer an AES-128 challenge, to ISO/IEC 29167-10 and the Authenticate part of Gen2 v2.0. The keys are held in the chip's guarded memory. A blank tag carrying the same EPC has no key to answer with.

But nothing happens by itself. The reader or the middleware has to send the challenge. A plain inventory read doesn't. If you buy authenticating tags and never ask them to authenticate, you've bought an expensive Gen2 tag.

## The copy you can't see

Here's where I'd push back on most of what's written about RFID security, including some of mine.

The TID is set at the factory and, on the chips I looked at, locked. NXP's DNA has a 96-bit TID with a 48-bit serial number. So a blank tag you buy keeps its own TID, and a back end that records the EPC/TID pair at encoding will see a mismatch. Add a last-seen location per tag and a number that appears in two cities in an hour stands out. This catches the ordinary copy and it costs almost nothing.

It doesn't catch all of them. The 2017 keynote lists the TID as a basic uniqueness check with no defence against emulators and none against chips with a writeable TID. It describes authentic EPC/TID data copied into a clone with a tag emulator. How common it is, I couldn't find a source for, so I won't guess.

> If the item is worth that effort, check the key, not the number.

## The reader

All of the above is about a tag telling a stranger a number. A reader can tell a stranger a lot more.

Many readers are small computers on your network. Impinj's own blog says the R700 runs a Linux kernel. It has a web interface, and readers like it speak LLRP, which IANA registers on TCP port 5084. A Nedap integration guide lists the R700's default login as root and impinj, and says the web interface has prompted for a change since firmware 9.0. Before that, nothing made you.

None of that is exotic. It's the same hygiene as any other box that talks to your ERP. Put the readers on their own segment, reachable from the middleware host and nothing else. Change the default logins. Keep them off the internet. Keep the firmware on a supported release. Then treat every read as untrusted input, because one is: a tag decides what it says.

The page has these as an eight-line checklist whose ticks live in the URL, so you can send your version to IT.

## What I don't know

I can't tell you your risk. It depends on what's on the tag, what a copy would be worth and who has a reason to walk your aisle. The matrix on the page is my reading of the GS1 standard, two vendors' published material and one conference keynote, and a security reviewer should read every cell before you rely on it.

Three things I'd check with a vendor before signing: whether the tag you're buying supports Protected Mode or authentication at all, whether your readers and middleware will use it, and who holds the passwords.

The aisle is the easy part to picture. The reader's login page is the one nobody opens.
