Back to SecuraCV

Unflashed boards

What protects you, and what can't · October 27, 2026

We tell you to buy a bare ESP32 and flash it yourself: commodity silicon, no vendor cloud. The catch: it arrives running a stranger's firmware. Here's what that costs and what to do.

The short version. Hold BOOT while you plug in a board you didn't flash yourself. Let our flasher wipe the whole chip the first time it meets it. Read what it tells you about the security fuses. And know that none of this covers a board that was physically tampered with.

The one move that matters

The rule

Never plug an unflashed board into your computer without holding BOOT.

  1. Unplug the board, if it's plugged in.
  2. Press and hold the BOOT button (usually marked B).
  3. Still holding it, plug the USB-C cable in. Then let go.

The chip comes up in its built-in download mode and the firmware it shipped with never runs a single instruction. That's enforced by ROM burned in at the factory, not by us.

Why this move: these chips have native USB, so their firmware decides what your computer sees. A hostile image can pose as a keyboard and type commands the instant it's plugged in, with no prompt and nothing built into Windows to stop it. In download mode the ROM presents only a serial port, so that attack is gone.

Why not just open our app first?

Because by the time any app or web page sees a USB device, your operating system has already set it up. Whatever the board wanted to do on plug-in has happened. Holding BOOT works earlier, inside the chip.

What an unflashed board can actually do

ConcernReal?What answers it
Types at your computer the moment you plug it inYesHolding BOOT. Nothing else does.
Joins open Wi-Fi, shouts on the airYesHolding BOOT — and it has none of your passwords yet
Hostile firmware survives your installYesWiping the whole chip first, which our flasher now forces
Flash chip is smaller than the label saysYesOur flasher reads both ends and compares
Board was locked by a previous ownerYesOur flasher reads the chip's security fuses and refuses it
Something extra soldered to the boardYesNothing. No software can see a modified circuit board.
Survives a full wipe and hides in your CanaryNoThe recovery ROM can't be written, and the device's identity key is created fresh after the wipe

What our flasher checks before it writes anything

Every one of these is a read. The flasher never burns a fuse, so a flash from the page can always be redone.

The chip's security fuses

The strongest signal, and free. A brand-new chip has every security fuse unset, so anything set was set by a person. If a previous owner turned on secure boot or flash encryption, the board is locked to their key for good: these fuses only burn one way. If any of the fatal ones are set, we tell you the board can't become a Canary before you waste an evening on it. If it was sold to you as new, it wasn't.

Whether the flash is the size it claims

A relabeled memory chip is a common marketplace fraud: sold as 16 MB, actually 4 MB, the top of the range mirroring the bottom. We read a block from each end and compare; on a chip that arrived erased, we say the check was inconclusive instead of guessing.

What it arrived running

We read the firmware's own name off the board. MicroPython, CircuitPython, Espressif's AT firmware, an Arduino sketch: all common, all fine. Something we don't recognize isn't an accusation, just worth a look before you erase it, and the flasher saves a copy either way.

Then it wipes the whole chip

A normal install writes the start of the chip: the bootloader, the partition table, a blank settings region and the firmware. It leaves the second firmware slot and any file system alone. That's fine for a Canary you already own (which still comes up with a new key, as after every USB install), and wrong for a stranger's board, because anything in the parts we don't write would survive. On a board our flasher has never seen, the full wipe is mandatory.

Both of our flashers read the fuses

The browser flasher reads the security fuses as part of its first look at a board. The desktop Flasher app reads them too, since version 0.12.0: open New board from a shop or marketplace? on step 1 and press Read its security fuses. The tool it drives has no command for that, so the app asks the chip's boot ROM itself, with read-only commands, and once it has read them a fatal fuse stops the flash just as it does in the browser. That native read is host-tested, not yet bench-tested on a board, and a read that can't finish says not checked, never “clean”, because a missing check reads exactly like a passed one.

The desktop app differs in three ways. Its fuse read ends by restarting the board into whatever it arrived running, which lets that firmware run — the thing holding BOOT was for. So on a board you didn't flash yourself, read the fuses with the browser flasher instead; if you already used the desktop read, unplug it and hold BOOT again before you flash (if the read found a fatal fuse, stop there: the app forgets the read once the board is unplugged). Its full wipe is a box, This board is new to me, ticked again for every board you plug in rather than forced: leave it ticked. And its Install a local file doesn't yet take that wipe or the safety copy, so it doesn't check the flash size either: for a board you didn't flash, use its normal install (Pick the firmware → Flash it).

The Lab app's own Flash page doesn't read the fuses or what the board is running, and takes no safety copy, so it doesn't check the flash size either. For a board you didn't flash yourself, use one of the two flashers.

What none of this protects you from

Stated plainly

A modified circuit board. An extra chip, a rewired USB path, a substituted module — nothing readable over a cable can see any of it. Buying from a distributor with real provenance, not a marketplace reseller, is the only answer, and a weak one.

Proving the silicon is genuine. There's no way, from your computer, to confirm a chip is really an Espressif part. A good enough clone answers every question correctly.

"Nothing looks wrong" is not "proven safe." It means the fuses are untouched, the flash is the size it claims, and the address is well-formed. That's a floor, not a guarantee, and we'd rather you hold it as one.

The opt-in lockdown, and why it isn't ready

The provisioning guide stages the same security fuses (secure boot and flash encryption), but as tooled today it turns on neither: it leaves flash encryption off and enables secure boot with nothing to verify against, and a board may not boot after it. So it does not yet make you the last person who can change what runs on the board. Even working, it would cover at most the firmware image: the identity key and your saved Wi-Fi sit unencrypted in the settings region on every Canary, so whoever holds the board can read them. The burn is permanent, which is why it is never a default.

Go deeper

The full engineering write-up — every fuse we read, where the numbers come from, and each check's failure mode: unflashed_board_intake.md

How we prove our firmware matches our public source: supply-chain transparency · the provenance trail

Found a hole in any of this? security.txt