Part · Faults & Errors

The infallible machine that isn't.

Camera fines arrive with the authority of a machine that "doesn't lie." But cameras have been hacked, mis-calibrated, and caught fining cars for speeds they can't physically reach — and the burden of proving the error is dumped on you.

When ransomware hit the cameras

In June 2017 the WannaCry ransomware infected Victoria's fixed road-safety camera network after a maintenance contractor plugged in an infected USB stick.

DetailFigure
Camera devices infected110 (43 Redflex + 67 Jenoptik)
Cameras potentially exposedup to 280
Fines suspended as a precaution~8,000
Fines withdrawn1,643
Fines embargoed pending review5,500

Authorities said there was no evidence the ransomware had altered any fine's accuracy — the tickets were cancelled "to preserve public confidence." Which rather concedes the point: public confidence in automated fines rests on the assumption the system is clean, and that assumption had just been broken.

Sources: iTnews; Global News; Victorian Road Safety Camera Commissioner report, WannaCry malware infection (2018).

Nearly 19,000 wrong fines — in one state, in three years

Sources: NSW refunds (documents released by the then-Opposition); Victoria faulty cameras / Datsun / West Gate. Some single-outlet specifics are reported figures; treat exact totals as "as reported."

The machine is not conclusive — a court said so

In RTA of NSW v Michell (2006), the NSW Supreme Court dismissed the roads authority's appeal against an acquittal. The camera image lacked a legally required security indicator (proof the image wasn't altered), and Justice Adams held this could ground reasonable doubt even with no contrary evidence from the driver:

Computer processes "can go awry," and a "tribunal of fact is still bound … to consider the quality of that evidence." — Adams J, RTA of NSW v Michell (2006)

In other words, a camera certificate is evidence — not automatic guilt. But most people never test that, because of how the system is built (see below, and The Law).

Source: RTA v Michell (2006) NSWSC.

The AI that "sees" a phone — and sometimes a sandwich

Mobile-phone detection cameras use AI to flag drivers, with a human reviewing before a fine issues. Reported grounds for successful challenges include a phone sitting untouched, a reflection or object misread as a phone, adjusting the air vents — even a hand near the face. The vendor's "95% capture" figure is a design target, not an independently verified false-positive rate.

The transparency hole Here's what's genuinely damning: no Australian authority publishes a false-positive rate or the proportion of contested camera fines that get withdrawn. NSW even hosts a dataset on fine reviews and waivers — but the headline error rate isn't advertised. Combined with owner-onus (you're presumed liable and must actively contest), the rational move for most people is to just pay. A system that profits from people not checking its accuracy has little reason to publish it.

Sources: NSW review/waiver data — data.nsw.gov.au; AAP FactCheck (enhanced images admissible). No official false-positive rate exists in public sources — a prime FOI/GIPA target.