The network that many alarm communicators still sit on is being retired. Let’s get you started on preparing your systems…

The country is retiring 2G one area at a time, and there's no deadline to work back from — here's how to find out whether your installed base is exposed…

Alarm communicators are among the most 2G-dependent devices in the country. So you'd think SA would have a solid plan for switching off that band.

Government was going to take care of it by 31 December 2027, but last month they decided not to, handing the entire process to the cell operators. So it's a commercial call now: Each operator switches off what it likes, when it suits their commercial and technical plans, and there's no national date to work back from.

Which means this won't happen at once. It'll happen area by area.

Nobody's switching 2G off wholesale any time soon, because too much still runs on it. It just shrinks instead, so a communicator won't die on a known date. It'll degrade, failing some nights and not others. Except where it doesn't degrade at all: Olarm is seeing whole towns and suburbs lose 2G effectively overnight.

And it's going exactly how you'd think: The only big company to say anything publicly is Santam, warning that many alarm and tracking systems still run on 2G and that legacy towers are already going dark.

The commercial problem is clear: When a device loses its network, it can stop working without the user knowing, and, of course, alarms and trackers are written into many insurance policy terms, so a customer's contents cover can depend on a working alarm, motor cover on a working tracker.

This is why an area-by-area shutdown lands harder on alarm companies than on anyone else. A tracker moves, so losing coverage in one suburb just means picking up a tower down the road. A communicator is bolted to a wall. If the 2G at that address goes and it can't reach a second tower from where it stands, it's offline, and nobody at that site is going to notice.

Dual SIM can make that worse by giving you false confidence: A unit can sit there happily on Vodacom while its MTN path died months ago. You find out the night Vodacom goes too.

So "online" isn't enough information. You want to know which network each device is actually on, whether that's 2G or 4G, its signal quality, and when it last successfully used its alternative network.

Start with the units already offline. Then the 2G-only ones with poor signal, because if their tower goes, they haven't got the reach to find another. Then the ones that haven't touched a second network in months.

Then don't dig the hole deeper: No more 2G-only installs, upgrade the old ones while a technician is already on site, and don't put units recovered from cancelled customers back into circulation.

When you do upgrade, buy modern LTE support and check the bands. At the 2022 spectrum auction, Vodacom took 700MHz while MTN and Telkom took 800MHz. In module terms, that's Band 28 and Band 20, and those are the bands the future runs on. They're also the low ones, so they travel further and punch through more, which is what you want on a farm far from a tower or in a basement. If B28 and B20 aren't on the spec sheet, keep looking. Add an independent path while you're there, WiFi or Ethernet, because two paths that fail together are one path.

And be honest about the maths: Thousands of customers means thousands of phone calls, appointments, site visits and installations, and at a handful a day per technician, this takes years. Put dedicated people on it, work suburb by suburb, and package the upgrade with a system inspection, a maintenance visit or app access so it isn't a cost conversation.

An Australian man handed the job of booking an oversubscribed gym class to an AI agent. It found the gym's booking limits existed only in the website, not the API behind it, so it booked him months ahead of what was allowed, then reported that the system ran no authorisation checks on cancelling other people's reservations (it already tested that on the person in waitlist position one). Asked to undo it, it could not. The flaw only ran one way.

Signal: Think about what runs on booking-and-cancellation logic in your world. Visitor pre-authorisation, contractor access bookings, gate permissions for a date range, delivery slots. Take one system you installed and check whether user A can cancel user B's entry by changing an ID in the request.

A new US system called SignalTrace works alongside plate readers, detecting the Bluetooth and RFID signals that devices already broadcast. It can recognise groups of devices that regularly travel together, and can tie those clusters to plate records and timestamps. The pattern is searchable without knowing the plate. The company behind it says the system does not identify people and only collects electronic signatures. For now, we assume.

Signal: "We don't collect personal information" is a claim about which fields sit in a database, not about what the system can do. Under the draft POPIA gated-access code, the estate is the responsible party, and you are the operator acting on instruction, so it pays to know what your kit collects.

A port authority detected a breach late on 4 August that took out the systems behind gate processing at three of its facilities, forcing a switch to manual check-ins to keep trucks moving. Gates were back on a normal schedule within two days. No data appears to have been taken, and nobody has claimed the attack.

Signal: Access control is an availability problem, not only a security one. Nothing touched the physical gates; the IT behind them just went away. If your client's gate system dies at 06:00 on a Monday, what is the written procedure, and who has it on paper?

SA’s Electronic Travel Authorisation system went live at OR Tambo this week, combining biometric verification, machine learning and an upgraded movement control system. During the pilot, more than 203 000 applications were processed and over 5 700 rejected, flagged by machine learning for fraudulent passports and manipulated documents. It expands to more countries next month and all visa types by year-end.

Signal: The state is normalising biometric identity checks at scale, which drags the standard for private access control up with it. Clients who clear a border by facial match will start asking why the estate gate still runs on a code from 2019.

A run of British venues has moved to bar Meta's camera glasses, mostly on the basis that filming staff or other customers without permission already breaches house rules. Europe's data-protection regulators are drafting guidance, and Germany's commissioner has suggested the glasses may already fall foul of hidden-recording law. It’s a potential privacy nightmare, and it might come to SA sooner rather than later.

Signal: A new category of thing is arriving at the door. If you run access control at a venue, clubhouse or corporate reception, write the policy before a client asks you for it.

Every alarm client you already service is a video client you have not sold yet. The Olarm APP can stream live camera footage and play back recordings alongside the alarm itself, which means video verification stops being a separate CCTV project and becomes an upgrade on a system that is already on the wall.

Three reasons it is worth your team's hour:

Get the integration right the first time and the upsell sells itself.

Security researcher Bill Swearingen spent a year and roughly 31 million tests training a model to generate patterns that surveillance cameras simply cannot see. It defeated all 11 open-source detection algorithms he tested, including the software behind Flock plate readers, Axon body cameras and Clearview AI, and it now turns out new patterns every minute, each batch mathematically better than the last.

On 7 August at DEF CON, he ran the first real-world test: a 2009 Toyota Yaris wrapped in one of the patterns, driven past a Flock camera. It worked, though the wheels were a problem. His strongest patterns stay offline so camera makers cannot train against them.

Signal: This failure mode produces silence, not an error. The camera records, nothing triggers, and there is no fault for anyone to investigate. Zero events overnight on a line-crossing analytic is not proof of a quiet night. Worth remembering that a pattern which beats vision still does not beat a PIR, a beam, a fence sensor or a radio.

Something on land and something on the water:

Running or fishing something we should be shouting about? Hit reply and put it on our radar.