Project Aurora Turns Online Speech Into Police Intelligence

Project Aurora is being sold as prevention. Its deeper significance is that Montreal police are moving further into a model where online language, social networks, and behavioural prediction become the raw material of criminal suspicion.

In March 2026, the Service de police de la Ville de Montréal announced Project Aurora, a new initiative aimed at detecting and preventing serious crimes allegedly ordered through social media. The SPVM described Aurora as an “innovative and disruptive” project using structured monitoring active day and night to intervene before violence targeting merchants takes place.

The public framing was straightforward: criminal networks had moved online, young people were being recruited through social media, and police needed to identify threats before shootings, arsons, and home invasions occurred. CityNews reported that Montreal police said 20 young people had been arrested through the initiative, with investigators analyzing coded language and sharing intelligence with other police forces.

By April 1, the SPVM announced four additional arrests connected to alleged online criminal contracts. Police said the investigations had support from the SPVM’s cyber-investigations unit and that suspects allegedly acted as sponsors or intermediaries in publishing contracts for violent crimes on digital platforms.

That is the official story: a targeted anti-violence operation aimed at serious crimes. But Aurora also represents something larger than one Montreal initiative. It belongs to a broader shift in Canadian policing away from traditional investigation of completed offences and toward continuous surveillance, behavioural interpretation, risk modelling, and pre-emptive intervention.

Aurora turns speech into intelligence

The central feature of Aurora is not simply that police monitor social media. Canadian police already do that. The more important shift is that Aurora treats online speech, coded vocabulary, social connections, and digital behaviour as intelligence material from which police infer future violence.

The SPVM says the project is focused on criminal contracts and serious violence. That limit matters. The concern is not that police should ignore credible threats. If someone is using a digital platform to arrange an arson, shooting, or home invasion, the public has an obvious interest in stopping it.

The danger begins when interpretation becomes infrastructure. Slang, emojis, jokes, threats, lyrics, memes, coded references, and online performance all become material for police analysts trying to distinguish noise from intent. That is an unstable process. Online language changes quickly. It is contextual, subcultural, ironic, exaggerated, and often deliberately ambiguous.

Once police assign criminal significance to unstable digital communication, evidence begins to merge with interpretation. A post becomes a signal. An association becomes a network. A joke becomes escalation. A digital persona becomes a risk profile.

Canada already built the model

Aurora should not be treated as an isolated innovation. Canada already has a documented history of police agencies using open-source intelligence tools to collect information from social media, forums, the dark web, location-based services, and private databases.

In 2024, the Office of the Privacy Commissioner released findings on the RCMP’s Project Wide Awake. The investigation found that since at least 2015, the RCMP had used private-sector services to collect personal information from open sources, including social media, forums, the dark web, location-based services, and fee-for-access databases.

The Privacy Commissioner’s findings matter because they show that Canadian policing has already normalized outsourced digital collection and open-source intelligence gathering. Aurora is more public-facing and local. Project Wide Awake was national and RCMP-based. The shared premise is the same: the digital traces people leave behind can be collected, organized, interpreted, and operationalized by police.

Citizen Lab’s report on algorithmic policing in Canada defines predictive policing as technology that processes mass data in hopes of predicting potential criminal activity before it occurs. The report distinguishes location-focused prediction from person-focused prediction, but both depend on the same basic move: convert historical and behavioural data into forecasts of future risk.

Aurora has not been publicly described as a fully algorithmic predictive-policing system. It should not be overstated as one without evidence. But it sits inside the same political and technical field: police using digital behaviour to identify risk before an offence is completed.

Mr. Big was behavioural engineering

The older Canadian model was not digital. It was theatrical. The “Mr. Big” operation became one of Canada’s most infamous undercover policing techniques: officers constructed a fake criminal organization around a suspect, built trust, offered money and belonging, then pressured the suspect to confess to an earlier crime before being fully accepted into the organization.

The Supreme Court of Canada described the dangers directly in R. v. Hart. The Court held that confessions produced through Mr. Big operations would be presumptively inadmissible, unless the Crown could show that their probative value outweighed their prejudicial effect. The Court recognized that the technique created serious reliability concerns and risks of abusive police conduct.

Hart mattered because it acknowledged something basic about police-created environments: people can be manipulated. A suspect can confess falsely because they want approval, money, protection, friendship, status, or escape from isolation. The confession may look voluntary while being produced by a world police built around the target.

In R. v. Mack, the Court applied the Hart framework and addressed the line between permissible undercover pressure and abusive state conduct. The point was not that every Mr. Big operation was unconstitutional. The point was that Canadian law had to confront the dangers of police manufacturing a social reality in order to extract incriminating behaviour.

That is the bridge to Aurora. The connection is not that Project Aurora is literally a Mr. Big operation. It is not, at least based on the public record. The connection is philosophical. Both models assume that police can enter or construct a suspect’s social environment, interpret the person through that environment, and intervene based on the behaviour that environment produces.

The digital version scales the logic

Traditional Mr. Big operations were labour-intensive. They required undercover officers, fake criminal networks, staged opportunities, months of contact, and a carefully engineered escalation of trust and pressure. Police had to build an artificial social world around one suspect.

Digital policing changes the scale. Online life already contains social worlds: friendships, rivalries, status hierarchies, grievances, financial desperation, performance, fantasy, recruitment, threat, and belonging. Police no longer need to build the environment from scratch. They can monitor environments that already exist.

That shift is enormous. Instead of one suspect inside one constructed criminal theatre, police can monitor many people across many online spaces. Instead of extracting a confession after a past crime, police can interpret behaviour as possible evidence of future risk.

This is where the criminal-law problem changes shape. Traditional criminal law punishes acts. Predictive policing watches probabilities. The subject is not judged only by what they have done. They are increasingly assessed through what police believe they may do, based on language, associations, affect, and pattern.

Entrapment law shows the pressure point

Canadian law has already had to confront police operations that create opportunities for crime rather than merely observe them. In R. v. Ahmad, the Supreme Court ruled that police need reasonable suspicion before offering someone an opportunity to sell drugs in dial-a-dope investigations.

The Court’s public summary stated the principle plainly: police need good reason to suspect the person answering the phone, or the phone number itself, is involved in drug dealing before asking them to sell drugs. That logic matters beyond drug cases. It reflects a basic constitutional anxiety: police cannot simply roam through social space offering opportunities for crime and then prosecute whoever responds.

That becomes more complicated online. Digital spaces make it easier to blur the line between observing risk and producing it. Fake accounts, undercover personas, monitored channels, automated flagging, and police-initiated conversations can all exist inside the same environment. The more police move upstream from completed offences, the more important the line between detection and manufacture becomes.

Aurora’s public materials do not establish entrapment. They do not show that police are inducing crimes. But they do raise the question that Canadian courts will eventually have to answer more often: when police enter digital spaces organized around suspicion, how much state participation is too much before prevention becomes production?

Racial profiling is already in the data

Predictive systems inherit the world that produced their data. In Montreal, that world includes documented racial profiling by police. This is not an abstract concern. CityNews reported on a 2023 update to research on SPVM street checks showing that Indigenous people were 4.6 times more likely to be stopped than white people, Black people 4.2 times more likely, and Arab residents twice as likely.

In 2024, a Quebec judge ruled that racial profiling was a systemic problem within the Montreal police force and awarded damages in a class action lawsuit. The ruling did not describe profiling as a hypothetical risk. It treated it as an existing institutional reality.

That reality matters for Aurora because surveillance does not begin on a blank slate. If certain communities are already more likely to be stopped, watched, documented, searched, and interpreted as suspicious, then intelligence systems built from police attention can reproduce that imbalance. More surveillance produces more data. More data produces more suspicion. More suspicion justifies more surveillance.

This is the feedback loop at the heart of predictive policing. A community is over-policed. The over-policing generates records. The records are treated as evidence of risk. The risk justifies continued over-policing.

Category drift is the political danger

The SPVM presents Aurora as a response to violent crime. That is the strongest version of the case for it. But surveillance infrastructures rarely remain confined to the emergency that justified their creation. Once police build a system for monitoring coded language, mapping networks, and identifying risk, the institutional pressure is to expand its use.

The categories can shift. Organized crime becomes gangs. Gangs become extremism. Extremism becomes public order. Public order becomes protest. Any community that uses encrypted communication, decentralized organization, anonymous participation, militant rhetoric, or distrust of institutions can begin to resemble a risk network when viewed through an intelligence lens.

Canada’s history makes that danger concrete. Indigenous movements, labour organizers, anti-globalization activists, environmental defenders, anti-war groups, and Palestine solidarity movements have all been treated as objects of security attention at different moments. The problem is not that Aurora is already being used against political dissent. The problem is that its logic can travel there easily.

That is category drift: the slow expansion of tools built for one purpose into adjacent fields of state concern. It rarely arrives as a dramatic announcement. It arrives as common sense. The same method works. The same analysts can use it. The same data is available. The same language of prevention applies.

The goal is behavioural management

The most powerful surveillance systems do not need to arrest everyone. They change behaviour by making people aware that they may be watched, interpreted, mapped, and flagged. Once people believe their language can be misread, their jokes can be recorded, their associations can be mapped, and their identity can become a risk signal, they begin regulating themselves.

This is where the Foucault reference matters without needing to turn the article into theory. The point of modern surveillance is not only visibility. It is normalization. People adjust conduct in anticipation of being observed. The police gaze does not merely collect information. It produces disciplined subjects.

Aurora and Mr. Big belong to the same historical trajectory because both move policing beyond evidence collection. Mr. Big built a fake criminal world around a suspect and used that world to extract truth. Aurora monitors digital worlds that already exist and uses them to infer risk before the criminal process has fully begun.

The technologies differ. The underlying logic does not. Canadian policing is moving from the investigation of acts toward the management of behaviour. Project Aurora is not the endpoint of that shift. It is one more sign that the future of policing is being built inside the spaces where people speak, perform, organize, joke, threaten, belong, and become legible to the state.

Sources