AI road safety cameras drive surge in driver fines

AI-assisted road safety cameras installed in multiple Australian jurisdictions have produced large numbers of automatic infringement notices. According to ABC News, Western Australia issued about 36,000 seatbelt infringement notices after the enforcement phase began in October 2025, with penalties starting at $550 and 4 demerit points per offence. News.com.au reports WA systems logged about 75,000 seatbelt and mobile-phone violations by mid-February 2026, and that the Victorian government collected more than $26 million in six months after introducing its own AI-powered cameras. The OECD AI incidents monitor lists the WA deployments as an AI incident, noting license losses and economic harms tied to camera-detected offences.

Per reporting in The Conversation, these systems use computer-vision models to screen still images from roadside cameras, flag potential offences, then pass flagged images for human review before issuing an infringement. That workflow is reported as the deployed model-plus-human-in-the-loop approach in the public rollouts. ABC News and other outlets document that cameras were configured to detect seatbelt misuse and mobile-phone use from still images captured on highways.

Industry-pattern observations: computer-vision enforcement systems commonly operate on high-threshold classifiers tuned to minimise missed offences while relying on human review to reduce false positives. In practice, edge cases such as children, partially occluded seatbelts, brief belt slips, atypical seating postures, or neurodivergent behaviours increase classification difficulty. Systems trained on limited datasets or lacking diverse edge-case examples can systematically overflag these instances, producing high-volume administrative burdens during appeals processes.

public reporting emphasises two fault lines. First, the technology delivers scale, enabling tens of thousands of processed images and automated flagging that translate directly into fines and demerit points, a material legal and economic impact for motorists (ABC News; News.com.au). Second, the current appeal and review pipelines are stressed by volume; ABC News documents community backlash and a formal review ordered by the WA Road Safety Commissioner after multiple contested fines, including instances where drivers were fined for passenger behaviour. The OECD monitor frames this as an AI incident because the system's outputs produced realised harms (fines, licence losses) and generated fairness concerns.

Editorial analysis: practitioners building or evaluating similar enforcement systems should note that detection accuracy on rare but consequential edge cases often matters more than aggregate precision metrics. High false-positive rates produce downstream social and legal friction, and operational costs for manual review can rapidly escalate. Public trust and perceived procedural fairness become as important as model performance when outputs have legal consequences.

• •Whether the WA review changes appeals processes or other procedures, as reported by ABC News.
• •Published performance metrics or independent audits of camera accuracy across demographic and situational subgroups.
• •Legal or regulatory responses in other Australian states after reporting of fines and licence impacts (News.com.au; OECD).
• •Any official statements or transparency releases showing model training data composition, false-positive rates, and human reviewer procedures.

the deployments demonstrate both the operational reach of vision-based enforcement and the governance challenges that follow when automated outputs carry legal penalties. Public reporting shows significant volumes of infringements and mounting community concern; observers should track audit, review, and regulatory outcomes for precedent on how jurisdictions balance automated enforcement, accuracy, and procedural fairness.