How a Single GPS Receiver Sent Telstra's Network Back to 2006

On 8 July 2026, a large part of the mobile network run by Australian operator Telstra stopped working: calls and text messages failed to go through, and trains, payment terminals, ticketing systems and electric-vehicle chargers all malfunctioned. Even calls to emergency-services numbers wouldn't connect. And yet no one attacked the network, cut the fibre or pulled the power. The cause, as an independent report by Technology Audit Partners revealed, was a single GPS receiver in one chassis in Melbourne that, after routine maintenance, decided the year was 2006.

The incident was dissected by Netnod, the Swedish operator that distributes the country's precise time. Its conclusion is unexpectedly instructive: modern cellular networks depend critically on a shared understanding of what "now" is. The 5G bands (including Sweden's main one at 3.5 GHz) operate in TDD mode, where transmission and reception alternate within the same block of spectrum. If one cell's clock "drifts," it will start transmitting during a neighbour's receive window — and the network will jam itself. The dependence on time here is a deliberate engineering choice.

How the error built up

Time at Telstra was distributed over the NTP protocol with a hierarchy of "strata." NTP has two safeguards against a bad source: priority for the lower stratum, and voting, in which a deviating source is discarded by the majority. But both had been quietly disabled — by two decisions taken five years apart, each of which looked sensible on its own.

In 2020, during an equipment replacement, the network moved from a strict client-server model to peering, where nodes negotiate the source among themselves. This opened the door to "time loops" — a situation where sources appear independent but in fact take their time from one another, like a rumour confirmed by three people who each heard it from the same person. In October 2025, engineers, tired of constantly losing the connection to the Sydney source, switched on an idle GPS receiver in Melbourne. No one realised that this turned an ordinary stratum-three server into a stratum-one server — that is, the most authoritative source for the country's largest mobile network.

The second ingredient is a well-known property of GPS: the week counter rolls over every 1,024 weeks (about 19.6 years). A running receiver simply keeps counting, but a switched-off one loses track of the epoch. The firmware on the Melbourne card hadn't been updated in six years, and on startup it threw the date back 1,024 weeks — to November 2006. Priority gave the floor to precisely that server, and there was no one to object: its partner server had been switched off back in 2020, and the remaining sources sat lower in the hierarchy and dutifully repeated the error. The further the wrong date spread, the more convincing it looked.

Takeaways

The protocol worked exactly as intended — it was the architecture around it that failed. Netnod advises treating time distribution as critical infrastructure: keeping documentation and "reference" configurations under version control, setting up round-the-clock monitoring, tracking vendor bulletins, building redundancy out of genuinely independent sources (rather than nodes that negotiate among themselves), and preferring dedicated connections over peering. The Swedes also single out Telstra for praise for commissioning and publishing an independent review — something everyone benefits from.

Source: Lobsters