IoT and kiosk breakout testing

A kiosk in a reception and a camera on a gate are the same problem: a computer sitting somewhere you do not control, usually on a network that trusts it more than it deserves.

Two problems, one skill set

Most estates have a category of device nobody owns. It was installed by a contractor, it runs an operating system nobody patches, and it sits on a network segment that was supposed to be isolated and is not. Visitor sign-in tablets, self-service terminals, ANPR cameras on the barrier, the access control panel in the riser, the building management controller in the plant room. All of them are computers, all of them are reachable, and almost none of them appear on the asset list that your last penetration test was scoped against.

Two kinds of testing cover this ground. Kiosk breakout asks whether somebody standing in front of a locked-down terminal can escape it and reach the operating system underneath. IoT and device testing asks what a connected device gives up when somebody has it on a bench, on the network, or in their hands. They use different techniques and they answer different questions, but they tend to arrive together because the estates that have one have the other.

What a kiosk breakout test actually does

A kiosk is an ordinary computer with a lid on it. The test is whether the lid comes off. We work from the position any member of the public has, standing at the device with no credentials and no tools, and then from the position of somebody who brought a keyboard.

  • Escaping the shell. Browser kiosk mode, a locked-down launcher or a full-screen application. The exits are rarely where the installer expected: a file upload dialog that gives you an Explorer window, a print dialog that reaches the filesystem, a help link that opens a full browser, an error page with a working address bar.
  • Keyboard and shortcut paths. Sticky keys, accessibility tools, task manager, the Windows key and its many combinations, function keys, and the debug shortcuts the vendor left in. A surprising number of kiosks are one key combination away from a desktop.
  • URI and protocol handlers. Links that open a mail client, a PDF reader, a remote desktop session or a local file path, each of which is a different application with its own file dialogs.
  • Physical and peripheral routes. Exposed USB ports, a keyboard emulator on a lanyard, a card reader that accepts more than cards, a touchscreen that behaves differently under a long press, the service hatch that was not locked.
  • What the escape is worth. Reaching a desktop is the beginning, not the finding. The finding is what that desktop can reach: cached credentials, a mapped drive, a domain join, a management VLAN, a database connection string in a config file, or the rest of the estate.

The last point is the one that decides severity. A kiosk that breaks out to a locked-down local account on an isolated VLAN is a low finding. A kiosk that breaks out to a domain-joined machine with a service account cached on it is the start of a very bad week, and we have found both.

The terminals this applies to

If a member of the public, a visitor or a contractor can touch it, it is in scope for this kind of test.

  • Visitor sign-in and reception tablets, including the ones that print badges
  • Self-service ordering, ticketing and payment terminals
  • Patient check-in screens and ward terminals
  • Library, college and school public access machines
  • Parcel lockers and collection points
  • Digital signage and wayfinding screens, which are usually a media player with a network port
  • Electric vehicle charge points with a screen
  • Time and attendance clocks on the factory floor

IoT testing: three layers, not one

A connected device is rarely one target. Testing it properly means treating it as three, because a weakness in any of them compromises the product.

What gets tested, and what each layer tends to give up
LayerWhat we doWhat it commonly yields
The device itselfExposed services, default and hardcoded credentials, debug interfaces, update mechanism, physical ports, storage at restShared credentials across every unit shipped, an open management port, an unauthenticated firmware upload
Firmware and configurationExtracting and reading the firmware where we can obtain it, looking for keys, certificates, credentials and the logic behind the authenticationPrivate keys committed to the image, a backdoor support account, signing that is not actually checked
The app and the API behind itThe mobile or web application that controls the device, and the cloud API both talk toOne account able to control another customer's device, which is usually the most serious finding of the three

That third row is worth dwelling on. The most damaging IoT findings we report are rarely exotic hardware attacks. They are ordinary authorisation flaws in the API, where changing a device identifier in a request reaches somebody else's camera, lock or meter. That is a web application and API testing problem wearing a hardware costume, and it is found by the same discipline.

Smart devices worth putting in scope

When organisations ask what counts, the honest answer is anything with an IP address that nobody logs into day to day. The ones we are asked about most:

  • ANPR cameras and barrier controllers. Number plate recognition sits on car parks, depots, secure sites and council schemes. The camera is a networked computer, the recognition feed is personal data under UK GDPR because a number plate identifies a keeper, and the controller usually talks to a barrier that somebody would quite like to open.
  • Access control and door entry. Panels, readers and the server behind them. Card cloning is the famous attack; the duller and more useful one is the management interface on the network with the installer's password still on it.
  • CCTV, NVRs and video management. Historically the worst-maintained devices on any estate, and the ones with the most direct privacy consequence when they fall.
  • Building management, HVAC and energy monitoring. Often installed by a facilities contractor, often reachable from the corporate network, occasionally reachable from the internet.
  • Smart locks, lockers and asset cabinets.
  • Printers and multifunction devices. Address books, scanned document caches, stored credentials for scan-to-folder, and an administrative interface that is rarely touched.
  • Industrial sensors, telemetry and remote monitoring kit, where the question is usually what the device can reach rather than what it holds.
  • Meeting room equipment. Screens, conferencing bars and room booking panels, all of which live on the network and listen.

What the UK rules require, and who they actually bind

The UK product security regime under the Product Security and Telecommunications Infrastructure Act came into effect on 29 April 2024 and is enforced by the Office for Product Safety and Standards. It sets three requirements for relevant consumer connectable products: no universal default or easily guessable passwords, published information on how to report security issues, and published information on minimum security update periods.

The part that gets misread is who it applies to. The duties sit with manufacturers, importers and distributors of relevant connectable products. If you buy and operate devices rather than make or sell them, the regime places no product security duty on you. That does not leave you with nothing to do: it means your obligation runs through your own risk management and your contracts with suppliers rather than through PSTI, and the useful question becomes whether the products you are buying meet it.

Several categories sit outside the regime entirely, including medical devices, smart meter products, electric vehicle charge points, and desktops, laptops and tablets without cellular capability. For a technical baseline to test against, the relevant standard is ETSI EN 303 645, the European baseline for consumer IoT security, which is what the UK requirements were drawn from.

If you build connected products, that changes the shape of the engagement: you are testing against a standard before a product ships, which is a different exercise from testing an estate that already has devices on it. We do both and they are scoped differently.

How this gets scoped

Three questions decide the shape and the price of a device engagement, and they are worth settling before anybody quotes.

  1. Do we get a unit to keep? Hardware testing is destructive in the sense that enclosures get opened and storage gets read. A spare unit we can take apart produces a far better test than a live one we must hand back working. If no spare exists, we test what can be tested without damage and say so in the report.
  2. How many device types, not how many devices? Four hundred identical cameras are one target. Four cameras from four vendors are four. Count models, firmware versions and configurations, not units.
  3. What does physical access look like? Public-facing kiosks are usually tested in place, in hours, with somebody from your side present. Devices in plant rooms and risers need an escort and a window. Neither is difficult, but both need arranging before the day rather than on it.

Where devices sit on a network you already have tested, this folds naturally into infrastructure and network testing rather than being bought separately, and that is usually the cheaper route.

What a device engagement costs

A kiosk breakout test on a single build is normally two to three days, which puts it at the lower end of our range. A device assessment covering one product across its hardware, firmware and API is typically four to eight days depending on how far into the hardware you want us to go. Mixed estates are quoted on the number of distinct device types rather than the number of units. What a penetration test costs sets out the UK ranges and the scoping drivers behind them.

There is no rush premium, and if a smaller scope would answer your actual question we will tell you that before quoting the larger one.

What you get

The same report as any other engagement: an executive summary somebody non-technical can act on, technical detail with full reproduction steps and evidence, severity rated against your environment rather than a generic scale, and remediation guidance your installer or integrator can work from. Critical findings reach you the moment they are confirmed. A retest is included once you have made the fixes.

For device work we add one thing: where a finding is the vendor's rather than yours, we write it so you can send it to them, and we will support a coordinated disclosure to the manufacturer if you want one. A good deal of what we find in this space is a product defect rather than a configuration mistake, and that is work we do anyway.

Talk to us about testing

This is a manual penetration test, not a vulnerability scan. Tell us what you need looked at and we will tell you what the work actually involves, what it costs and when we can do it. The testing is done by the person you speak to.

Solusec is CREST accredited as a company, and every test is carried out by a qualified, CREST registered tester. Certifications held include CREST CRT, OSCP, OSWE, CISSP and CISM, alongside registration with the UK Cyber Security Council.

Your details are handled by a real person, never fed into AI.

Common questions

What is a kiosk breakout test?

A test of whether somebody standing at a locked-down terminal can escape the application it is running and reach the operating system underneath. We work first with no tools and no credentials, as any member of the public would, then with a keyboard and the techniques an attacker would bring. The finding is not the escape itself but what the escape reaches: cached credentials, mapped drives, a domain join or the rest of your network.

Do you need to take the device apart?

Only if you want the hardware layer covered, and only if you can give us a spare unit. Opening an enclosure and reading storage directly is the most productive part of a device test, but it is not reversible in any way we would want to promise on a live unit. Without a spare we test the network, application and API layers and state plainly in the report what was out of reach.

Can you test ANPR cameras and barrier systems?

Yes, and they are a common request. An ANPR installation is a networked camera, a recognition service and a controller attached to something physical, so the test covers the device, what it exposes on the network, who can reach the management interface, and whether the barrier can be operated by somebody who should not be able to. The recognition feed is personal data under UK GDPR, because a number plate identifies a registered keeper, so how it is stored and who can read it is part of the scope.

Does the PSTI Act apply to us if we only operate the devices?

No. The duties under the UK product security regime fall on manufacturers, importers and distributors of relevant connectable products, not on organisations that buy and operate them. If you are an operator, your obligations run through your own risk management and your supplier contracts instead. The practical use of the regime for you is as a purchasing question: ask whether the products you are buying meet it.

Can you test in place, or does equipment have to come off-site?

Both work. Public-facing kiosks are usually tested in place during opening hours with somebody from your side present, because that is the realistic attacker position anyway. Hardware teardown needs a bench and a spare unit, so that part comes to us. Devices in plant rooms and risers are tested on site with an escort and an agreed window.

Will the testing break anything?

On a live estate we work to avoid it, and the rules of engagement set out what must not be touched and what happens if something goes wrong. Device testing carries more risk of disruption than a web application test, because some of these devices do not handle unexpected input gracefully, which is itself a finding. Where a test is likely to disrupt, we say so first and you decide whether to run it in hours, out of hours, or on a spare.

Do you test the mobile app and cloud API as well as the device?

Yes, and that is usually where the serious findings are. The most damaging IoT issues we report are ordinarily authorisation flaws in the API, where changing a device identifier reaches somebody else's equipment. Testing the hardware and ignoring the API covers the interesting part and misses the dangerous one.

How many devices do you need for a quote?

Count device types, not units. Four hundred identical cameras on one firmware version are a single target. Four different models from four vendors are four targets and four times the work. Send a list of models, firmware versions and roughly where each sits on the network and we can scope from that.

How is this different from a normal network penetration test?

A network test finds these devices and reports what they expose from the network. A device test goes further: physical interfaces, firmware, stored credentials and the application behind it. If you have never tested the estate at all, start with the network test, because it tells you what is actually out there. Device testing is the depth pass on what that finds.

Can you test building management, HVAC and access control?

Yes. These are among the most commonly neglected systems on an estate because they were installed by a facilities contractor rather than by IT, and they frequently sit on a network segment with more reach than anyone intended. Scoping them needs the integrator involved early, since the useful test often requires access they control rather than access you do.

Is device and kiosk work covered by your CREST accreditation?

Solusec is CREST accredited as a company and the work is delivered by qualified, CREST registered testers, with the same methodology, reporting and quality assurance as any other engagement we run. Testing is scoped and authorised under written rules of engagement in the same way.

Related

Ready to talk?

Scoping conversations are free and there is no sales team to get past. Tell us what you're dealing with and we'll tell you honestly what you need.