What if CCTV cameras could think? SA company bets on AI to fight crime

A Cape Town company believes it may have found a way to help security teams stop crimes before they happen by teaching CCTV cameras how to think, adapted for local situations.

Using artificial intelligence (AI), Safeza’s Swiss-developed AVA-X surveillance software pledges to scan thousands of hours of video footage within seconds, recognising faces and objects, tracking suspicious behaviour in real-time and alerting security teams before incidents escalate.

In a country where violent crime dominates daily life and investigators often spend days combing through grainy CCTV footage, the promise of faster, smarter policing is attracting attention locally and internationally.

However, experts warn that while AI surveillance may strengthen investigations and improve response times, it could also deepen privacy concerns, reinforce racial bias and create dangerous overreliance on technology if not properly regulated.

Safeza AVA-X (Pty) Ltd founder and CEO Armand de Beer said the company was born out of frustration with South Africa’s crime crisis and outdated security systems.

“South Africa has one of the highest crime rates in the world, and the truth is our current security systems are mostly reactive — they record what happened, but they don’t help prevent it,” said De Beer. “Investigators are spending hours, sometimes days, manually reviewing CCTV footage to find a single face or vehicle. We want to change that.”

The biggest difference is that our system runs entirely on your own infrastructure. Your data never leaves your premises, never goes to a cloud server, and is never accessible to us or any third party

—  Armand de Beer, Safeza founder and CEO

The company’s name reflects that mission. Safeza stands for “Safe South Africa”. De Beer described the technology as “turning standard CCTV cameras into intelligence”.

Its software is divided into three systems:

• Sentinel Investigation allows investigators to search archived CCTV footage almost instantly using a face, object or appearance description.
• Sentinel Live monitors live camera feeds and flags unusual activity or known suspects in real-time.
• Sentinel Access uses facial recognition to control entry into secure spaces.

To explain how the technology could work in practice, De Beer used the example of a shopping centre repeatedly targeted by armed robbers.

“Investigators can upload a description or image of a suspect into our Sentinel Investigation platform and search the entire archived CCTV library in seconds rather than days,” he said.

“Meanwhile, Sentinel Live is monitoring all the live camera feeds in real-time. If someone matching a flagged profile enters the premises, the system alerts the security team immediately so they can respond.”

The company said its technology differs from many global AI surveillance systems because it runs entirely on-site instead of relying on cloud servers.

“The biggest difference is that our system runs entirely on your own infrastructure. Your data never leaves your premises, never goes to a cloud server, and is never accessible to us or any third party,” said De Beer.

“We also don’t rely on OpenAI, Google or any big-tech platform. Our AI models were developed independently in Switzerland and have been validated in real operational environments, including European law enforcement.”

International recognition

The start-up recently gained international recognition after being selected to present at the SelectUSA Investment Summit in Washington DC, hosted by the US department of commerce. Out of more than 230 companies globally, Safeza was selected among the top 48 finalists and went on to win the SelectTech defence category.

“In that room, we pitched in front of US governors, senators and major investors alongside eight other innovative defence companies,” said De Beer. “For a South African start-up, that is a remarkable validation.”

The technology has been used in one of the world’s most high-profile missing persons investigations. According to De Beer, the company’s Swiss technology partner assisted investigators after a Polish woman claimed she was Madeleine McCann, the British child who disappeared in Portugal in 2007. Using facial recognition analysis, the system concluded it was “practically impossible” for the woman to be McCann.

“The system distinguished cleanly between two separate questions, which is exactly what you need from a forensic-grade tool,” said De Beer. “That result was covered internationally and remains one of the clearest public demonstrations of what our technology can do in a real-world, high-stakes missing persons context.”

For many South Africans living with the daily fear of hijackings, robberies and violent attacks, the appeal of such technology is understandable. Private security companies play a growing role in communities where policing resources are stretched thin and AI-powered surveillance is increasingly being seen as a possible force multiplier.

Our vision is for technology built and proven in South Africa’s hardest conditions to become a genuine tool for making South African communities safer

—  Sparkly Mokgosi, Safeza senior advisor

De Beer believes the technology could fundamentally change the way security teams operate.

“Real-time anomaly detection means security teams can act before a crime is committed, not only after,” he said. “For communities that are most vulnerable to violent crime, that shift from reactive to proactive policing could be genuinely life-changing.”

Safeza, which is looking to secure a commercial rollout, told TimesLIVE the technology has been adapted for South African conditions.

“The Swiss-engineered foundation we started with is excellent, but South Africa has specific challenges: extreme lighting conditions, high-density environments and crime typologies you don’t encounter in Europe,” said its senior advisor Sparkly Mokgosi.

“We are building our local deployments specifically to train the AI on South African data. Our goal is to build the world’s most stress-tested AI video intelligence system and South Africa, with all its complexity and difficulty, is exactly the right environment to do that.”

“We are also working across road safety, retail, critical infrastructure and private security, so the impact isn’t limited to one sector. Our vision is for technology built and proven in South Africa’s hardest conditions to become a genuine tool for making South African communities safer.”

Experts weigh in on using AI to fight crime

Prof Mpho Primus, co-director of the Institute for Artificial Intelligent Systems at the University of Johannesburg, warned against viewing AI as a cure for crime.

“Crime is not simply a pattern-recognition problem,” he said. “It is tied to inequality, unemployment, institutional weakness, infrastructure failure and social fragmentation.”

Primus said AI systems can improve efficiency by helping security teams process large amounts of information quickly and identify unusual patterns. However, technology alone cannot fix broken systems.

“If the underlying institutional systems are weak, AI simply scales those weaknesses,” he said. “There is also a misconception that more surveillance automatically equals more safety. It does not.”

He said facial recognition systems carry serious ethical risks because they can inherit biases from the data used to train them.

“These systems are not neutral,” said Primus. “In African contexts, this concern becomes even more serious because African populations remain under represented in many global AI datasets.”

Dr Nicholas Bacci, senior lecturer in anatomical science at the University of the Witwatersrand, agreed AI surveillance has strong potential, but warned mistakes could have devastating human consequences.

I often consider how powerless one would feel in that situation, effectively facing a system with a purported 99% accuracy rate, where being part of the remaining 1% could have life-altering consequences

—  Dr Nicholas Bacci, Wits University

“The critical issue is not only system performance, but how video analytics outputs are interpreted, verified by and integrated in rigorous human-led decision-making processes,” said Bacci.

He warned that people wrongly identified by facial recognition systems could face life-changing consequences.

“The major ethical concern, in my opinion, is the risk of false accusations, arrests, imprisonment and similar harms. I often consider how powerless one would feel in that situation, effectively facing a system with a purported 99% accuracy rate, where being part of the remaining 1% could have life-altering consequences.”

Legal experts have also questioned whether South Africa is fully prepared for widespread AI-driven surveillance.

Simon Dippenaar & Associates said South Africa lacks a strong national facial image database linked to criminal records.

“At the moment, the SA Police Service lacks a robust, centralised database of mugshots linked to criminal records, so even if a criminal is caught on camera, there may not be a match,” the firm said.

The firm also warned about privacy risks and racial bias. “Facial recognition software has been shown to incorrectly identify people of colour,” it said.

Despite the concerns, the government appears increasingly interested in AI-driven policing tools. During the 2025 state of the nation address, President Cyril Ramaphosa said South Africa planned to adopt “surveillance, analytics and smart policing solutions for modern law enforcement”.

Mokgosi, asked about concerns with this technology, said: “Facial recognition and AI surveillance are powerful tools, and like any powerful tool, they carry real risks if misused, bias, false positives, erosion of privacy, or deployment without proper oversight. These are legitimate concerns and we take them seriously.

“Our response to them is structural: human-in-the-loop oversight means a person is always in the decision chain. Full auditability means every action can be reviewed and challenged. POPIA compliance means the legal framework for data protection applies to every deployment. We are not building a system that operates in the dark. We are building one that can be scrutinised, questioned, and held accountable.”

For Safeza, the goal goes beyond international recognition or commercial success. “Our core philosophy says it in the name,” said De Beer. “We want to protect our communities, our friends, our families in South Africa.”

TimesLIVE

---