Cellular IoT security usually gets written about as a list of things to worry about. I want to make the opposite case, carefully. In my piece on Anthropic’s self-replicating AI research, I said in passing that cellular estates have controls most broadband sites can only dream of. The more I dug into it, the bigger that point got. Used properly, the SIM and the network behind it can be your best early-warning system, your quarantine switch and your asset register, all at once. The catch is the word “properly”. Too much of the installed base is still running on public IP SIMs, default passwords and firmware nobody has touched since installation day. With AI worms on the horizon, botnets of hijacked routers already here and new UK and EU rules landing, now is the time to fix it. It will cost something. Not fixing it will cost a great deal more.
The short answer: cellular IoT can be one of the most defensible ways to connect a device, because the network itself can isolate it, lock it to its hardware, cap and watch its traffic, and cut it off in seconds. Providers such as Onomondo, emnify, Eseye and Vodafone IoT already offer these controls. But they only help if they’re switched on and the device isn’t sitting on a public IP with an open admin page. The installed base needs a review now: start with what your SIM platform already tells you, make the remote changes first, and fold the rest into site visits you’ll be making anyway.
First, let’s agree on the threat
None of this works unless we’re all on the same page about what we’re defending against. Here’s where things stand in late 2026:
- Hijacked routers and IoT are now infrastructure for attackers. In April, the NCSC and international partners published an advisory on China-nexus covert networks, built mainly from compromised small office routers plus IoT and smart devices. Attackers route their activity through them to hide.
- Old edge kit is the soft spot. In February, US agency CISA ordered federal bodies to find and replace end-of-support edge devices within 12 to 18 months, and explicitly included IoT edge devices and routers. It published supporting guidance jointly with the FBI and the UK’s NCSC.
- AI is about to make it worse. Anthropic’s Frontier Red Team thinks small AI models may be able to copy themselves across cheap, badly secured IoT devices by the end of this year. They’d be slow and hard to shut down, and they’d adapt to hardware they’ve never seen.
- The money is real. IBM’s 2026 report puts the average UK data breach at £3.13 million, and it takes UK organisations an average of 165 days just to spot one. The Jaguar Land Rover attack in 2025, which stopped production rather than stealing data, cost the UK economy an estimated £1.9 billion and hit more than 5,000 organisations, according to the Cyber Monitoring Centre.
Most of that has nothing to do with cellular. But a lot of the kit in those botnets looks very like the routers and gateways this industry has been installing in cabinets, plant rooms and kiosks for twenty years.
Where cellular can shine, serve and protect
Here’s the bit that doesn’t get said loudly enough. On broadband or a customer’s Wi-Fi, the network is someone else’s problem and you see almost nothing. On cellular, the network is a managed service with a control panel. Every packet passes through a core network that your provider runs, and most of the serious IoT connectivity providers now let you use that core as a security tool.
| Network control | What it stops | Examples on the market |
|---|---|---|
| Private IP addressing and private APNs | The device can’t be found or reached from the internet, so it can’t be recruited into a botnet from the internet | emnify says its devices are never given public, internet-routable addresses; Eseye builds around private APNs |
| SIM-to-device lock | A stolen SIM won’t work in other hardware | Onomondo restricts each SIM to approved hardware; emnify says a SIM moved to different hardware won’t attach |
| Isolation between SIMs | One compromised device can’t reach the rest of your fleet | Onomondo describes network-level isolation separating device profiles at the SIM |
| Usage caps and alerts | Spots and limits a device suddenly sending or pulling far more data than it should | Standard on most IoT platforms, including Onomondo’s usage limits and alerts |
| Traffic visibility | Lets you see exactly where a device is talking to, without touching it | Onomondo’s Traffic Monitor captures every packet to and from a device in the core network |
| Anomaly detection | Flags unusual behaviour, location or roaming automatically | Vodafone IoT uses machine learning to flag unusual device behaviour; Eseye partners with Armis for threat detection |
| Suspend by API | Quarantines a suspect device in seconds, from anywhere | Available on most modern connectivity platforms |
Those are examples, not endorsements, and features vary by plan. But put that table next to Anthropic’s advice for defending against an AI worm: watch for the network signature, then quarantine infected hosts. In the research Anthropic cites, ISP quarantine cleared 92% of Mirai infections within 14 days. On a cellular estate, you can be your own ISP. A SIM on a few megabytes a month that suddenly pulls a multi-gigabyte model is about as loud an alarm as you’ll ever get, provided someone set the alert.
There’s a quieter benefit too. Your SIM platform already knows where every connected device is, which network it’s on, when it last attached and how much data it uses. For a lot of organisations, that’s the closest thing they have to an accurate asset register for their remote kit. The NCSC and CISA both start their edge device advice with “know what you’ve got”. On cellular, you mostly already do.
Why you haven’t seen it put together like this
Every one of those features exists today. What I haven’t seen is anyone line them up side by side, set them against the actual threats, and tell buyers what to ask for. The reason is simple: the people who talk about these controls are the people selling them. Each provider makes the case for its own platform, in its own language, on its own website, and the bigger point gets lost in the marketing. Buyers end up comparing price per megabyte, because that’s the one number everyone publishes.
That’s the gap IoTPortal is here to fill. We’re independent, we don’t sell SIMs, and we can say plainly what each control does, where it falls short and which questions sort the serious providers from the rest. If this is useful, pass it on to whoever buys your connectivity, whoever installs your routers and whoever signs off your security budget. It’s the conversation they should be having.
Where it doesn’t, and why honesty matters
I’m not going to pretend the SIM is a magic shield. Cellular has had its share of overselling, and there are people in security who argue its defences stop short in OT environments. They have a point. The limits are real:
- A public IP SIM undoes most of it. If the router has an internet-facing IP and an admin page open to the world, the network controls above are watching an attack instead of preventing one. I’ve written about how public IP SIMs became a security liability, and PHP2Shell shows what happens to exposed admin pages.
- The SIM sees the site, not the LAN. Behind a router, the network sees one SIM’s traffic. It can’t tell you whether the camera or the signage player sent it.
- Features are often off by default, or extra. Private APNs, alerts and IMEI locks only protect you if somebody switches them on and somebody reads the alerts.
- It doesn’t patch anything. A router with a known flaw, like the recent Teltonika RutOS CVE, still needs its firmware updated. The network buys you time and visibility, not immunity.
Cellular’s advantage is control and visibility, not invulnerability. If we’re honest about that, buyers will believe the rest.
The installed base is the real problem
New deployments are the easy bit. You can spec private APNs, decent routers and proper credentials from day one. The hard bit is everything already out there: years of routers installed with public IP SIMs so somebody could reach a web page remotely, admin passwords never changed from the label, firmware that was current the day it went in, and hardware that’s since gone end-of-life. I’ve seen what that looks like in practice in the supermarket freezer hack: OT kit lives far longer than the security assumptions it was installed with.
Nobody is going to budget a fleet-wide site visit just for security. The good news is they don’t have to. A lot of the fixes can be done remotely, and the rest can ride on visits that are happening anyway: maintenance calls, network migrations as 2G heads for switch-off, hardware swaps and contract renewals.
Why now
Every year this gets put off, it gets dearer. Several things have come together this year:
- The EU Cyber Resilience Act’s reporting duties are live. Since 11 September 2026, manufacturers must report actively exploited vulnerabilities in their products, including products already on the market, with an early warning within 24 hours. The full CRA follows in December 2027. If you sell connected kit into the EU, your installed base is now a reporting liability.
- The UK’s Cyber Security and Resilience Bill is on its way. It brings medium and large managed service providers into scope for the first time and lets regulators designate critical suppliers. If you manage connectivity or remote kit for essential services, expect questions.
- The attackers have industrialised. The NCSC advisory describes Chinese security companies building and maintaining covert networks for multiple state-linked groups. Old routers are a supply chain for them.
- AI changes the maths. Anthropic’s point was that lightweight AI agents may be far more willing than human attackers to grind through the world’s cheap IoT, one device at a time.
- Insurance won’t cover what you didn’t do. JLR reportedly had no cyber cover in place when it was hit. Insurers are asking harder questions about basic controls every renewal.
The high price of protection, and the higher price of not
Let’s be straight: securing an installed base costs money. Moving SIMs to a private APN, replacing end-of-life routers, updating firmware and paying someone to watch alerts all come with a bill. But compare like with like. IBM’s average UK breach is £3.13 million. Supply chain compromise was the single biggest cost driver in IBM’s UK figures, and the Cyber Monitoring Centre says future high-impact events are more likely to be disruptive attacks than data theft. That’s exactly what a hijacked OT estate looks like.
Set against that, a configuration change pushed from a connectivity platform costs almost nothing. A firmware update done during a planned visit costs a bit more. A replacement router at a site you were visiting anyway is a line item. The expensive option is the one where you find out what your estate was doing from a regulator, an insurer or the NCSC.
A practical plan for the installed base
- Inventory from the SIM up. Export every active SIM from your connectivity platform: where it is, what network, last attach, data use. Match it against what you think is installed. The gaps are your first findings.
- Do the remote quick wins. Set usage caps and alerts on every SIM. Turn on IMEI or hardware locks. Restrict destinations where your platform allows it. None of this needs a site visit.
- Kill the public IPs. Move to a private APN or VPN-based remote access. If something genuinely needs inbound access, put it behind a VPN, not an open web page.
- Fix credentials and firmware. Change default passwords, close admin access from the WAN and update firmware, remotely where your router platform supports it.
- Replace end-of-life kit on the next visit. Anything out of vendor support goes on a replacement list and gets swapped when someone is next on site, not when it fails.
- Decide who watches. Alerts nobody reads are worthless. Agree who owns them: you, your installer or your provider. That’s the whole theme of who secures the router in the cabinet.
Questions to ask your IoT SIM provider
- Do my SIMs get private IP addresses by default, and is a private APN included or extra?
- Can you lock each SIM to its device, and alert me if it moves to different hardware or a different country?
- What alerts can I set on data use, roaming and destinations, and how quickly do they fire?
- Can I restrict each SIM to a list of destinations my devices are allowed to talk to?
- Can you capture or show me a single SIM’s traffic without touching the device?
- Is your anomaly detection about security, or only about network performance? Who looks at it, you or me?
- Can I suspend or quarantine a SIM instantly via API or the portal?
- Can you stop SIMs in my fleet from reaching each other?
- Can you give me a full export of my SIMs with location, network and last activity, so I can build an asset register?
A good provider will answer most of these with a yes and a demo. If yours can’t, that tells you something too.
My take
I’ve spent a long time defending cellular against the charge that it’s expensive and fiddly next to Wi-Fi or a LoRaWAN gateway. On security, I think we’ve undersold it. No other way of connecting a remote device gives you a managed network that can isolate it, watch it, cap it and switch it off from a browser. That’s a genuine silver lining, and as an industry we should be shouting about it.
But we only earn the right to shout if we clean up our own back yard first. The same industry that sells private APNs also installed thousands of routers on public IPs with the password on a sticker. The threats coming over the next couple of years, from state-run botnets to AI that spreads itself, will find that kit. The question is whether we find it first.
Share this with your team. If you buy, install or manage cellular IoT, send this to the person who chooses your SIM provider and the person who owns your security risk, and use the questions above at your next contract review. IoTPortal will keep tracking these threats and the controls that answer them, so subscribe to the newsletter if you want the next piece first.
Quick answers
Is cellular IoT more secure than Wi-Fi or LoRaWAN?
Not automatically. Cellular’s advantage is that the network is a managed service with security controls: private IP addressing, SIM-to-device locks, isolation, usage alerts, traffic visibility and instant suspension. Those controls make it one of the most defensible options, but only when they’re switched on and devices aren’t exposed on public IPs.
What security features should an IoT SIM provider offer?
Look for private IPs or private APNs, IMEI or hardware locking, isolation between SIMs, usage caps and alerts, destination restrictions, traffic visibility, anomaly detection, instant suspension via API and a full SIM export for your asset register.
How do I secure an existing cellular IoT estate?
Start with an inventory from your SIM platform, then set alerts, caps and SIM locks remotely. Move devices off public IPs onto a private APN or VPN, change default passwords, update firmware, and replace end-of-life hardware on the next scheduled site visit.
Why is now the time to secure installed IoT devices?
EU Cyber Resilience Act reporting duties have applied since 11 September 2026, including to products already on the market. The UK Cyber Security and Resilience Bill is bringing managed service providers into scope. The NCSC has warned that state-linked attackers are using botnets of compromised routers and IoT, and Anthropic expects small AI models to be able to self-replicate on cheap IoT by the end of 2026.
How much does a cyber attack cost a UK organisation?
IBM’s 2026 report puts the average UK data breach at £3.13 million. Disruptive attacks can cost far more: the Cyber Monitoring Centre estimates the 2025 Jaguar Land Rover attack cost the UK economy about £1.9 billion.
Sources: NCSC and partners, “Defending against China-nexus covert networks of compromised devices”, April 2026; CISA Binding Operational Directive 26-02 and joint CISA/FBI/NCSC edge device guidance, February 2026; Anthropic Frontier Red Team, “AI models may be capable of self-replicating without access to a datacenter”, 12 October 2026; IBM Cost of a Data Breach Report 2026, UK edition; Cyber Monitoring Centre assessment of the Jaguar Land Rover incident, October 2025; European Commission, CRA reporting obligations; UK Cyber Security and Resilience (Network and Information Systems) Bill; provider documentation from Onomondo, emnify, Eseye and Vodafone IoT. Provider features are as published by each company and vary by plan.



