Toronto’s new AI police cameras: safety upgrade or step toward big brother?

Artificial intelligence cameras that can read license plates and recognize vehicles are no longer science fiction in Toronto. They are already mounted in a handful of locations across the city, quietly scanning traffic in real time. Supporters say they could help solve serious crimes faster. Critics worry they are another step toward a "big brother" style surveillance system.

What are these AI police cameras and how do they work?

The cameras Toronto police are using are a newer generation of automatic license plate readers (ALPRs). These systems capture images of passing vehicles and use AI to extract useful details automatically.

Earlier versions of this technology have existed in Ontario since the late 1990s, starting with systems that monitored vehicles entering and exiting Highway 407. Later, police began installing ALPRs in patrol cars, especially in traffic units, to quickly check plates against databases.

The latest cameras go a step further. In addition to reading the letters and numbers on a plate, they can also recognize and categorize details about the vehicle itself, such as make and model. That means the system can match not just a plate number, but the type of car it’s attached to, making it easier to flag suspicious or wanted vehicles.

Why are these cameras being deployed now?

Toronto police say they currently have five of these AI-enabled cameras operating in the city. According to their public statements, the cameras are installed only in areas with a recent history of violent activity and are used to identify vehicles tied to potential criminal activity.

There’s also a broader policy backdrop. When Ontario removed annual license plate registration fees, the government funded more ALPRs so police could still track whether plates were properly registered. Over time, that infrastructure and funding helped normalize the use of automated plate-reading technology in policing.

Policing as an “information game”

Modern policing increasingly revolves around information. The more data police have about people and places, the more they can map patterns, track suspects, and respond to incidents. These AI cameras fit into a long trend of expanding police data collection.

Earlier examples in Ontario include controversial practices like carding (street checks), where officers collected personal information from people who were not suspected of a crime. Other tools include body-worn cameras and fixed surveillance cameras. AI-enhanced license plate readers are the latest addition to this ecosystem, automating and scaling up what used to be manual observation and note-taking.

Key concerns: surveillance, bias, and “dragnet” policing

Civil liberties groups and many researchers worry that these cameras could turn into a form of dragnet surveillance, especially if they are concentrated in certain neighborhoods. When police place powerful monitoring tools only in specific areas, it raises questions about who is being watched most closely and why.

There is a history in Canada of policing practices disproportionately affecting racialized communities, including with carding. Critics fear that AI cameras could quietly recreate similar patterns under a more high-tech label: constant monitoring in some neighborhoods, far less in others.

Beyond location bias, there’s a broader civil liberties concern. If people feel that every movement in public space is being tracked and stored, they may be less willing to protest, gather, or simply move freely. Over time, that can chill free expression and association, even if most of the data is never actively used in an investigation.

When the AI gets it wrong

Another major worry is accuracy. AI-driven plate readers are not perfect. Misreads can and do happen, and the consequences can be serious.

In the United States, there have been documented cases where an ALPR flagged the wrong vehicle. In one incident, police were looking for a stolen motorcycle but stopped a minivan with a family inside because the plate characters matched what the system thought it saw. The family ended up in handcuffs before officers realized the mistake.

These errors are not just technical glitches; they become real-world police actions. And research suggests that when officers see a computer-generated alert, they may treat it as nearly infallible. Instead of using their own judgment to question whether the alert makes sense, they may default to “the computer must be right,” even when there are obvious red flags.

Can safeguards and audits prevent abuse?

Toronto police say they have safeguards in place. According to their public statements, access to camera data is tightly controlled, included in training, and limited to designated members on a need-to-know basis. They also say that every access attempt is logged and can be audited.

On paper, that sounds reassuring. But recent history in Ontario raises doubts about whether logging and auditability are enough. There have been multiple cases where officers allegedly misused police databases for personal reasons, including stalking women they knew or had seen in public places like gyms. In Toronto, there have also been controversies over unauthorized access and leaks of sensitive information, including body-worn camera footage.

These systems were also supposed to be auditable and controlled. Yet the public has seen little follow-through in terms of discipline or transparent consequences. That track record makes it harder for people to simply trust that new AI systems will be handled differently.

How long is the data kept?

Data retention is one of the few concrete details Toronto police have shared. For these AI cameras, they say:

• Ordinary “read records” (routine scans that don’t match anything on a watchlist) are stored for 7 days before being purged.
• “Hit records” (matches to a hot list, such as a stolen vehicle) are kept for 365 days and then deleted.

On the surface, this looks like a best-practice approach: keep routine data briefly, and hold onto relevant investigative data longer. However, this is only one part of the picture. The bigger, less visible question is what the technology vendors themselves do with the data.

The hidden role of AI vendors and data reuse

Many police technologies are supplied by private companies that also store and process the data. For example, Toronto uses Axon for body-worn cameras. In the United States, Axon has used footage from police forces to train AI tools that help automatically draft police notes.

That kind of secondary use is often not obvious to the public—and sometimes not fully understood by frontline officers either. With the new AI cameras, the vendor is reportedly a Canadian company based in Ottawa, which may help keep data within Canadian borders. But it still leaves open crucial questions: How long does the company keep raw footage? Is it used to train new AI models? Can it be shared with other clients?

As AI becomes more deeply embedded in public systems, these vendor practices matter just as much as official police policies. Similar issues are emerging across the AI ecosystem, from avatar tools to video generators, where training data and identity risks are hot topics—see, for example, concerns raised around YouTube’s AI avatar tools and identity headaches.

Do we need new rules and regulators?

There is growing recognition that existing oversight structures may not be enough for AI-driven policing tools. In Ontario, a new Inspector of Policing has been created to look at systemic issues like database access and information misuse, beyond individual misconduct cases. That investigation should eventually give the public a clearer picture of how widespread database abuse really is.

Still, many experts argue that we need more explicit laws and regulations at the provincial or federal level to govern how AI surveillance tools can be used. That could include:

• Clear limits on where and when AI cameras can be deployed
• Strict rules on data sharing between police and private vendors
• Mandatory transparency about error rates and misuse
• Strong penalties for unauthorized access or leaks

Other regions, especially in Europe, are already moving toward tighter controls on how governments and companies collect and use personal data. In the United States, AI-powered systems like Flock cameras have sparked intense political debates and even acts of civil disobedience, with some people physically removing or damaging the devices out of privacy concerns.

Are there real benefits to public safety?

Despite the risks, there are potential benefits. Recent investigations in Toronto have shown how valuable video evidence can be. In a high-profile shooting at a foreign consulate, police used conventional surveillance footage to quickly identify the vehicle involved and lay charges within a week.

AI cameras could, in theory, speed up that process by automatically logging every vehicle that passed the scene, cross-referencing plates and vehicle types against existing databases, and surfacing likely suspects faster. For serious violent crimes—especially those involving vehicles—shaving days off an investigation could matter.

The question is whether the marginal gains in speed and efficiency justify the long-term expansion of automated surveillance infrastructure, especially when many of these areas are already heavily covered by traditional cameras.

What should citizens do in an AI-surveilled city?

For everyday people, the first line of defense is knowledge. Understanding your rights when interacting with police, and your expectations of privacy in public spaces, is essential in a world where AI surveillance is becoming normal.

In Canada, you generally have the right not to answer police questions if you are not under arrest, and you can ask whether you are free to go. You also have constitutional protections around unreasonable search and seizure, freedom of expression, and freedom of association. But these rights only help if you know them well enough to assert them calmly in real situations.

That may mean reading up on the Charter of Rights and Freedoms, discussing these issues with friends and community groups, or following reporting and analysis on AI and policing. The same way people are learning how to use AI tools for productivity or content creation—like automating content workflows with systems such as Claude Code for SEO and writing—we also need to learn how to live with AI systems that watch, record, and sometimes misinterpret our actions.

The bigger conversation Toronto needs

Ultimately, the debate over AI police cameras is not just about one tool. It’s about what kind of city residents want to live in, and where they want to draw the line between safety and privacy.

Toronto’s experience shows how quickly new technologies can be adopted with minimal public consultation. AI cameras may be limited to five locations today, but without clear rules and active civic engagement, that number could quietly grow. The real safeguard is not just technical controls or internal policies—it is a well-informed public willing to question, debate, and, when necessary, push back.

As AI continues to move from science fiction into everyday infrastructure, that kind of democratic oversight will matter more than ever.