AI & Automation

Over-the-Air Firmware Updates for IoT Devices: How to Update a Fleet Without Bricking It

November 11, 2026 · 7 min read · Etzal Group Team

A water utility installs 400 pressure loggers across its network. A year later, a bug is found: after 255 days of uptime, the device clock wraps and timestamps go wrong. The fix is ten lines of code. The devices have no remote update. Each one must be visited, opened, connected to a laptop and reflashed: 400 site visits, some in chambers that need traffic control to open. The hardware had been chosen on price. Over-the-air firmware updates were listed as an optional extra and declined.

Why remote update is not optional

Every device ships with bugs nobody has found yet. Over a five to ten year service life:

  • Bugs appear that only show in the field: after months of uptime, at temperature extremes, with a particular network operator.
  • Security vulnerabilities are discovered in the device's software or the libraries it uses.
  • Networks change. Operators retire technologies, certificates expire, server addresses move.
  • Requirements change. A new measurement interval, a new alarm rule, a new data format.

Without remote update, each of these means a truck. With it, each means a controlled release from a desk.

Six requirements for a safe mechanism

1. Authenticity: the device only accepts genuine firmware

Each firmware image is digitally signed by the manufacturer or by you. The device holds the matching public key and refuses anything not correctly signed. Without this, anyone who can reach the device can install their own software on it.

Ask where the signing key is kept. It should be in protected storage, with few people able to use it.

2. Integrity: the download is complete and uncorrupted

The device checks a cryptographic hash of what it received before installing. A partial or damaged download is discarded.

3. Atomic installation: old or new, never half

The standard design uses two firmware slots, often called A/B. The device runs from one, writes the new version to the other, verifies it and switches on the next restart. Power lost during the download changes nothing: the running version was never touched.

Devices with only one slot overwrite themselves and are at risk during every update. Prefer two.

4. Automatic rollback: a bad version cannot strand the device

After switching, the new firmware must prove itself: start up, run a self-check and reach the server. Only then is it marked as good. If it fails to do so within a set time, a watchdog restarts the device into the previous version.

This single feature is the difference between a bad release and a bricked fleet. The test of "good" must include connectivity. Firmware that runs perfectly and cannot connect is the worst case, because you can never send it a fix.

5. Protection against downgrade

The device refuses versions older than a minimum, so an attacker cannot reinstall an old version with a known hole.

6. A protected channel

Updates travel over an encrypted, authenticated connection. Signing protects the content. The secure channel protects against tampering with which version is offered, and keeps the firmware itself confidential where that matters.

Constraints in the field

Real fleets have limits that shape the design.

  • Bandwidth. A device on a low-power wide-area network may be able to receive only a few hundred bytes at a time. Full images can take hours or be impractical. Delta updates, which send only the differences, reduce size greatly. On the most constrained networks, plan for very small, rare updates.
  • Battery. An update costs energy. A device meant to run five years on a battery cannot be updated monthly. Check the battery level before starting, and budget for a known number of updates over the device's life.
  • Intermittent links. Downloads must resume where they stopped.
  • Data cost. Four hundred devices times several megabytes over cellular adds up. Know the bill before the rollout.
  • Timing. Do not update a device while it is doing something that must not be interrupted. Let it choose a safe moment within a window.

Rolling out in stages

Never send a new version to the whole fleet at once.

Stage 0: the bench. Test on every hardware revision you have deployed, including the oldest. Test the update itself, from each version currently in the field to the new one, and test rollback by deliberately installing a faulty build.

Stage 1: a handful. Five to ten devices you can reach physically, in real field conditions. Watch for several days.

Stage 2: a small share. About 5 percent, spread across sites, hardware revisions and network conditions. Watch for a week.

Stage 3: a larger share. 25 to 50 percent.

Stage 4: the rest.

At each stage, check before continuing:

  • Did every device report the new version?
  • Did any roll back, and why?
  • Are they reporting data as before? Same interval, plausible values?
  • Battery drain, signal and restart counts: any change?
  • Any device gone silent?

Have a stop button. If something looks wrong, pause the rollout. Devices that have not updated yet are safe.

Knowing what you have

An update system needs an inventory.

  • Every device, its hardware revision and its current firmware version.
  • Last contact time.
  • Update history: what was offered, when, and the result.
  • Groups: by site, customer, hardware type.

Without it, you cannot answer "how many devices are still on the vulnerable version?", which is the first question after any security notice.

The devices that do not update

In every rollout a few devices will not take the update: offline, low battery, poor signal, a fault.

  • Keep offering it for weeks, not hours.
  • List the stragglers and find out why each one failed.
  • Decide what happens to those that never update: a site visit, replacement, or isolation if the update fixed a security problem.

Configuration is not firmware

Many changes do not need new firmware at all. Reporting interval, alarm thresholds and server address should be settings that can be changed remotely, separately from the software.

  • Faster, smaller and lower risk than a firmware update.
  • Still needs validation: a wrong server address strands a device as surely as bad firmware. Devices should fall back to the last working configuration if a new one stops them connecting.

Questions to ask a hardware supplier

Before buying devices in quantity:

  1. Does the device support remote firmware update? Over which networks?
  2. Is firmware signed, and who holds the signing key?
  3. Is there a dual-slot design with automatic rollback?
  4. What does the device check before marking new firmware as good?
  5. How large is an update, and are delta updates supported?
  6. What is the battery cost of one update?
  7. Can updates be rolled out to groups, in stages, and paused?
  8. Can we see each device's version and update history?
  9. For how many years will the supplier provide security updates, and what happens after?
  10. Can we update the devices ourselves if the supplier stops trading?

A supplier who cannot answer the fourth and ninth questions has not operated a fleet.

Who decides, and when

Agree in advance:

  • Who approves a release.
  • What testing is required before each stage.
  • When updates may run: not during a critical operating period.
  • Who is told: site owners, customers, the operations team.
  • How urgent security updates are handled, with a shorter path.

Regulation is moving in this direction as well. Product security rules in several markets now expect manufacturers of connected devices to provide security updates for a defined period. If you sell devices, check what applies where you sell.

Common mistakes

  • Buying devices without remote update to save on unit price.
  • Unsigned firmware.
  • A single slot with no rollback.
  • "Good" defined as "it started", without checking connectivity.
  • Updating the whole fleet at once.
  • Testing only on the newest hardware.
  • Ignoring battery and data cost.
  • No inventory of versions.
  • Forgetting the stragglers.
  • A signing key on one engineer's laptop.

We specify update mechanisms and run staged rollouts for the monitoring systems we build, as part of IoT and Control Systems. If you are choosing hardware for a deployment that must last years, contact us before the order is placed.

Summary

Treat remote update as a basic requirement of any connected device. A safe mechanism signs firmware, verifies downloads, installs into a second slot, rolls back automatically if the new version cannot start and reconnect, and refuses downgrades. Design for limited bandwidth and battery, roll out in stages with checks and a stop button, keep an inventory of versions, chase the devices that did not update, and put configuration changes on a separate, lighter path.

Frequently asked questions

What is an over-the-air firmware update?

It is replacing the software inside a device remotely, over its network connection, without anyone visiting it. The device downloads the new version, checks that it is genuine and complete, installs it and restarts. For a fleet of sensors spread over many sites, it is the only practical way to fix bugs and security problems.

What happens if an update fails halfway?

With a well-designed mechanism, nothing bad. The device keeps its previous working version in a separate area of memory, installs the new one beside it, and only switches over after verifying the download. If the new version does not start or cannot reconnect, the device returns to the old one automatically.

How often should IoT devices be updated?

As often as there is a reason: a security fix promptly, a bug fix when it affects operation, new features on a planned schedule. Many fleets settle on a few planned releases a year plus urgent security updates. What matters most is having the ability, tested, before it is needed.

Author

Etzal Group Team

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

WhatsApp AI Chatbot

We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.