MotorBuzz has been covering the global expansion of AI enforcement infrastructure across its Drivers Revenge section for months, from the New Zealand Acusensus contract that handed a private Australian company $100 million to film and algorithmically fine Kiwi drivers, to San Francisco's 369 per cent citation surge after automated enforcement went live, to California's push to fine the vehicle rather than the driver to improve collection rates. The pattern is consistent across every jurisdiction. The cameras multiply. The revenue grows. The accidents do not stop.
The defenders of automated enforcement are not lying when they say the research shows a safety benefit. A PLOS One study using Bayesian causal inference methods found a mean 15 per cent reduction in collisions at speed camera sites in England. The DfT's own evaluation of the Safety Camera Funding Scheme found a 22 per cent reduction in personal injury collisions at camera locations and a 42 per cent reduction in people killed or seriously injured. An LSE study in 2017 calculated that adding 1,000 cameras to UK roads could save up to 190 lives annually. You can cite those numbers honestly and they do not disappear.
But look at what those numbers are actually measuring. Collisions at camera sites. Not overall road safety. Not the road network as a whole. Cameras are placed at sites that already had an above average crash history, and then collisions at those specific sites reduce. Some of that reduction is genuine deterrence. Some of it is regression to the mean: sites with unusual concentrations of accidents tend to return toward average rates regardless of any intervention. Some of it is the well documented kangaroo effect, drivers braking sharply near fixed cameras and accelerating away afterward. A 2010 ICM Research report estimated that one per cent of all UK accidents are caused by drivers braking and then accelerating near speed cameras, equating to 28,000 accidents nationally per year. You cannot cite the collision reduction near cameras without also citing what happens on the road between them.
The deeper problem is structural. Speed cameras are reactive by design. The UK's national guidelines require cameras to be placed primarily at accident sites. That means a child has to have been hit before the camera goes up. The camera is a response to danger that already proved itself fatal or serious. Nobody argues for cameras at locations before a pattern of deaths is established, because without the historical collision data the placement cannot be justified. The logic of the system is: wait for the crash, document the crash, install a camera, reduce crashes at that specific spot, report the reduction as evidence of success.
What this system cannot do is prevent the first crash. What it cannot do is address a dangerous road design before anyone dies on it. What it cannot do is change the behaviour of a driver who has never encountered the camera before, or who drives distracted, drunk, impaired or deliberately reckless rather than merely exceeding a limit by a few miles per hour on an empty road at 3am.
Virginia recorded nearly one million camera triggers in 2025, as MotorBuzz reported in the Drivers Revenge section. A million interactions between automated enforcement and drivers in a single US state in a single year. If fines worked as a deterrent the way the industry claims, those numbers would fall sharply year on year as drivers learned and adjusted. In New York City, a 2025 study published in ScienceDirect found that camera effectiveness was not sustained beyond the initial months after installation, with some clusters showing a median increase in tickets by the fourth month. Drivers adapted to the cameras, not to safer driving. They learned where the cameras were and slowed for them specifically.
That distinction is the core of the argument. Slowing for a camera is not safer driving. It is performance. And a system that teaches performance rather than behaviour is a revenue system dressed in safety language.
Like this? Get the app: iOS | Android
The revenue framing is not a conspiracy theory. It is written into the architecture of the system. In the UK, the Safety Camera Funding Scheme that expanded the network from 2000 onward was explicitly structured so that partnerships of police and local authority agencies could recover the costs of operating cameras from the fines they generated. Income from fines funded more cameras. More cameras generated more fines. The scheme was not designed to make itself obsolete as roads became safer. It was designed to grow. When the Coalition government cut the Road Safety Grant in 2010 and decoupled camera funding from fine revenue, several local authorities switched cameras off or left them in an unfunded state. Half of UK speed cameras were not operational as of a 2017 Freedom of Information request. The ones that stayed on were overwhelmingly concentrated in areas where the traffic volumes and fine collection rates made them financially viable.
Now the same model is being handed to private companies and scaled up with AI. Acusensus in New Zealand is not a public safety body. It is an ASX-listed Australian company with shareholders, a board and a profit motive. It charges per fine processed. Its commercial interest is in maximum citation volume, not minimum accident rates. When MotorBuzz investigated how Kiwis are fighting mass AI surveillance on their roads, the fundamental question was not whether the cameras work but who benefits when they do. The answer is not the child about to cross the road. The answer is the algorithm, the company and the government collecting the revenue.
The argument that catches are not the same as prevention is simple enough that it survives the complexity of the research. We do not allow food safety inspectors to operate on a model where they are paid per fine issued and have no obligation to prevent contamination before it sickens anyone. We do not allow building inspectors to collect revenue from every structural failure they document after the roof has fallen in. We insist that prevention is the point. For road safety, somehow, we have accepted a system where the point is the catch.
If the goal were genuinely the child crossing the road, we would spend the camera money on raised crossings, reduced speed limits enforced by road design rather than by detection, better lighting, school zone engineering that makes 20 miles per hour the physically natural speed rather than the legally required one. The Netherlands has done this for decades. Its road death rate is among the lowest in the developed world. It did not achieve that by installing more cameras. It achieved it by making the roads themselves prevent the crash.
That is built-in obsolescence in the best possible sense. Infrastructure that solves the problem and then has no further role. The antithesis of a revenue model. Which is precisely why it is not what gets built.
Sources: PLOS One, "Do speed cameras reduce road traffic collisions?" September 2019 | DfT Safety Camera Funding Scheme evaluation data | LSE study, "The impacts of speed cameras on road accidents," 2017 | ScienceDirect, "Assessing the impact of fixed speed cameras on speeding behavior and crashes," March 2025 | Wikipedia / Road speed limit enforcement in the United Kingdom | ICM Research / LV Insurance speed camera report, 2010 | MotorBuzz Drivers Revenge | San Francisco speed cameras | Acusensus NZ | Virginia cameras | California vehicle fining