Cape Town’s AI traffic cameras showed promise. The law will decide

The City of Cape Town says it has no immediate plans to deploy artificial intelligence (AI) traffic cameras at scale, despite early success from a pilot project that used AI-powered cameras to detect motorists driving without seatbelts, using mobile phones behind the wheel, or committing offences such as crossing solid white lines.

Following the pilot, the city approached the National Director of Public Prosecutions (NDPP), South Africa’s top prosecutor, for guidance on whether evidence gathered through AI-enabled cameras would be admissible and legally enforceable.

“There is no plan for a mass rollout of AI traffic cameras; however, we cannot discount the possibility that AI will play a bigger role in this and other means of enforcement in the future,” Kevin Jacobs, spokesperson for Cape Town’s traffic services, told TechCabal.

Cape Town’s cautious approach reflects a broader challenge facing governments globally: balancing the efficiency of AI surveillance with legal and constitutional protections.

AI-powered enforcement tools are becoming common in countries such as China and parts of the European Union, where they support road safety and public surveillance systems. 

However, regulatory frameworks differ sharply across jurisdictions, shaping how the technology is used.

In South Africa, any expansion of AI traffic cameras would need to comply with the Protection of Personal Information Act (POPIA). While law does not ban the use of AI-powered images, it classifies facial images and biometric identifiers as “special personal information,” meaning authorities must demonstrate a lawful basis for collecting such data, limit collection to what is necessary, and ensure strong security safeguards. 

“Continuous road surveillance captures far more than just offenders. Law-abiding motorists, passengers, and pedestrians are also recorded. That makes purpose, minimisation, limitations, and retention rules critical,” said Nerushka Bowan, a technology and privacy lawyer and the founder of Law Innovation Technology Tomorrow (LITT) Institute.

Bowan noted that, in addition to the POPIA Act, for evidence collected through AI-powered cameras to be admissible, it must comply with the Criminal Procedure Act and established evidentiary principles. For example, the system must be reliable and accurate; there must be a clear chain of custody for the data; the evidence must be capable of being challenged and tested in court; a human decision-maker must ultimately exercise judgment.

If an accused person cannot meaningfully challenge how the AI reached its conclusion, that raises fairness concerns. South African constitutional law places a high value on procedural fairness and the right to challenge evidence.

Municipalities and prosecutors will need a technical understanding of how these AI systems work, have standards in place for accuracy and auditing, implement clear accountability lines when systems fail, and have the capacity to explain AI evidence in court. 

Cape Town has increasingly turned to technology to supplement its limited human enforcement capacity. Jacobs said enforcement agencies face mounting pressure as traffic violations continue to surge.

“Technology is becoming ever more critical in the enforcement space, as there simply is not the human resources to respond to the many demands on the City’s services,” he said.

The city currently operates approximately 77 static speed cameras, alongside mobile speed-trapping operations in high-risk areas.

According to Jacobs, the use of technology has already produced tangible enforcement outcomes. In one recent case, authorities used surveillance tools to arrest three suspects, including two taxi drivers with nearly 80 outstanding warrants worth more than R300,000 ($17,700).

The city recorded more than 2.7 million traffic offences in the previous financial year, although officials believe the actual number of violations is likely higher.

Distracted driving remains a persistent risk

South Africa banned texting or using handheld mobile phones while driving under the National Road Traffic Regulations in 2011. Cape Town has been among the most aggressive municipalities in enforcing these rules, with local by-laws allowing for the impoundment of mobile phones used illegally behind the wheel.

Over the past four years, however, Jacobs said the city has recorded a decline in cellphone impoundments and fines. He attributed this trend partly to increased compliance among drivers and the widespread adoption of hands-free technology in newer vehicles.

“The phenomenon of distracted driving continues to be a challenge, but the downturn in statistics does suggest that some drivers have heeded our appeals to not use their cellphones while driving,” Jacobs said.

Global regulation is diverging

Cape Town’s wait-and-see approach mirrors global regulatory fragmentation around AI surveillance. The European Union’s AI Act bans most forms of real-time facial recognition in public spaces, reflecting strong privacy protections. China’s 2025 regulations limit biometric surveillance in sensitive locations such as hotels and bathrooms, while still allowing extensive public monitoring.

The United States, by contrast, has no comprehensive federal AI surveillance law. Instead, states such as Illinois require explicit consent before collecting or using facial recognition data.

For Cape Town, the next step will depend largely on whether South African prosecutors consider AI-assisted traffic detection legally defensible. If approved, the technology could significantly expand automated traffic enforcement and help address resource constraints.

For now, the city is signalling that while AI may eventually become a central part of road safety enforcement, its large-scale deployment will hinge on legal clarity and public trust.

“Done carefully, AI-powered cameras can improve safety and consistency. Done poorly, they risk normalising opaque surveillance and weakens trust in enforcement systems,” said Bowan.