In this article
  1. What researchers have shown
  2. Habit as a digital fingerprint
  3. Why this is a concrete problem
  4. What this means for privacy protection
  5. Some practical ideas
  6. Table: what makes you recognizable, and what you can do about it
  7. Informed vigilance, not paranoia
  8. Frequently asked questions
  9. Further reading

Privacy policies often claim that your data is “completely anonymous”. Research in artificial intelligence shows that this claim deserves a much closer look.

Digital privacy isn’t a stable state. It is a moving balance that evolves with technological capabilities. Each time a new protection mechanism emerges, new methods to get around it appear a few years later.

That is exactly what happened with the notion of data anonymization. For a long time, removing names and a few direct identifiers from a dataset seemed enough to make it harmless. Research in artificial intelligence now seriously challenges that assumption.

What researchers have shown

Researchers trained an artificial intelligence system on a database of more than 40,000 phone numbers, along with general information about the people each of those numbers contacted. The goal was to determine whether the system could, from these behavioural patterns alone, identify an individual within a previously anonymized dataset.

The reported result: the system correctly picked out the target in more than 50% of cases. The method relied on a simple but effective logic.

The AI’s reasoning can be summed up like this: “This phone number regularly contacts these four other numbers. By cross-referencing existing data on those four contacts, the anonymous data point that interacts with them probably corresponds to this particular person.”

Computer scientist Jaideep Srivastava, of the University of Minnesota, commented on the work at the time, noting that it isn’t surprising that people tend to stay within established social structures, and that these regular interactions form a stable pattern over time.

In other words: humans are creatures of habit, and that is precisely what these systems exploit.

Habit as a digital fingerprint

The research focused on phone numbers, but the underlying principle applies to a much wider range of digital behaviour. A person’s habits form what could be called a behavioural fingerprint — a profile as distinctive as a signature, even with no name attached.

Social and communication habits

  • People contacted regularly
  • Frequency and times of calls
  • Social networks used
  • Groups and communities frequented

Browsing and consumption habits

  • Websites visited regularly
  • Usual browsing times
  • Online shopping habits
  • Apps used and how often

Location and travel habits

  • Regular routes
  • Travel times
  • Places visited repeatedly
  • Modes of transport used

Physiological and daily habits

  • Sleep and wake times
  • Physical activity patterns
  • Eating patterns
  • Use of connected devices

Taken separately, each of these elements seems harmless. Combined and analyzed over time, they form a behavioural profile precise enough to allow identification, even without any direct identifier.

Why this is a concrete problem

The theoretical framework is interesting. The practical implications are even more so. Here are some concrete situations where this kind of technological capability can have real consequences.

Situation 1 — Regular browsing: you regularly visit the same five sites at consistent times. Even with a VPN active, this behavioural pattern could potentially identify you from browsing metadata, without your real IP address being known.

Situation 2 — Everyday apps: an alarm clock app sends anonymous data to third parties. Your wake-up time is always the same. This signal, combined with other behavioural data, may be enough to re-identify you within a dataset that is supposed to be anonymous.

Situation 3 — Health data: information about a medical condition, initially treated as anonymous, can be re-identified and potentially sold to third parties — insurers, advertisers or others — without your knowledge.

Situation 4 — Refined ad targeting: even if you try to limit your exposure to advertising, systems able to re-identify you from anonymous data can build a precise profile of your behavioural vulnerabilities — spending habits, recurring interests, weak points — and exploit them commercially.

These situations don’t all require a single technology. In the most advanced cases, it is the combination of several complementary methods that produces precise results. AI re-identification is part of a broader ecosystem of behavioural analysis.

What this means for privacy protection

Data anonymization rests on an assumption: that removing direct identifiers makes information harmless. That assumption is increasingly tested by the current capabilities of AI-assisted behavioural analysis.

Re-identification: the process by which a person is identified from previously anonymized data, by cross-referencing it with other sources of information or analyzing characteristic behavioural patterns.

That doesn’t mean every precaution is useless. It means that protecting digital privacy calls for a more nuanced understanding of what is really protected — and what isn’t — when a privacy policy talks about anonymization.

Principle of digital vigilance: anonymizing data reduces some risks but doesn’t eliminate them. Understanding the limits of this protection lets you make better-informed decisions about what you share, with whom and in what context.

Some practical ideas

There is no universal, absolute solution to this kind of problem. Technical approaches evolve constantly, and what works today may work less well tomorrow. That said, some directions remain relevant for a reasoned reduction of exposure.

  • Question the apps you install. Every app collects data. Some sell it to third parties. Regularly assess which apps you really need, and read privacy policies before accepting — paying particular attention to clauses about sharing data with third parties.
  • Vary your digital habits. Stable behaviour is what makes re-identification possible. Occasionally changing your browsing times, tools or usage patterns makes your behavioural profile less readable, with no particular technical effort.
  • Limit the permissions you grant apps. Location, contacts, microphone, photo access: each permission granted is an additional source of data. Reducing these accesses to the strict minimum automatically shrinks your exposure.
  • Don’t overestimate the protection a VPN alone offers. A VPN hides your IP address but doesn’t change your behavioural patterns. It is a useful tool within a broader approach, without being a complete solution on its own.
  • Distinguish what is public from what is private. Some information shared in contexts you think are private — closed groups, forums, health apps — can end up in datasets that are sold. This distinction is worth making consciously before each share.

These directions aim to reduce exposure, not to guarantee absolute anonymity. In citizen preparedness, the goal is to understand the real risks in order to make informed decisions — not to achieve an illusory protection based on false certainty.

Table: what makes you recognizable, and what you can do about it

Data sourceWhat it revealsReasonable reduction
Phone locationHome, work, routes, places you visit“While using the app” access only, location history turned off
Contacts and callsYour social networkDeny contact access to apps that don’t need it
BrowsingInterests, schedules, habitsA browser that blocks trackers, third-party cookies refused
Health apps and connected devicesSleep, activity, medical conditionsRead the sharing clauses, turn off unnecessary syncing
Online shopping and loyalty programsConsumption habits, family situationLimit accounts, refuse sharing with partners

Informed vigilance, not paranoia

Digital privacy is a fast-moving subject. The capabilities of AI systems in behavioural analysis keep advancing, and regulatory frameworks often struggle to keep up.

In this context, the most useful stance is neither indifference nor anxiety. It is active understanding: knowing what is collected, understanding how it can be used, and adjusting your practices accordingly — without pretending perfect solutions exist.

Institutional frameworks generally provide legal protections around the use of personal data — notably Law 25 in Quebec, or the GDPR in Europe. These frameworks are a useful reference, but they don’t necessarily cover every real use of anonymized data, especially when that data is resold to third parties operating in other jurisdictions.

Knowing the limits of these protections is part of an informed approach to self-reliance.

This article covers a fast-changing subject. The capabilities of AI systems in re-identification have continued to advance since the research cited. Regular monitoring of the subject is recommended for anyone who wants to keep an up-to-date understanding of the real risks.

Frequently asked questions

Is using a VPN enough to protect my anonymity?

A VPN hides your IP address and encrypts your connection, which is a useful layer of protection. However, it doesn’t change your behavioural habits — the sites you visit, the times, the usage patterns. These elements can be enough to re-identify you regardless of your IP address. A VPN is one tool among others, not a complete solution.

Does truly anonymous data exist?

In theory, yes. In practice, the line between anonymous and re-identifiable data depends on the analytical capabilities available at a given time. What was technically anonymous ten years ago may no longer be today. Anonymization reduces risks but doesn’t eliminate them for good.

What does Quebec law say about data protection?

The Act to modernize legislative provisions as regards the protection of personal information, commonly called Law 25, governs the collection, use and disclosure of personal information in Quebec. It notably imposes transparency and consent obligations. However, it doesn’t necessarily cover every case of re-identification from anonymized data, especially when third parties operate outside Quebec. Checking the current texts is recommended for any specific application.

Should I delete all my apps to protect myself?

No. The goal isn’t to erase all digital presence, but to adopt a conscious stance toward what you share and with whom. Regularly reviewing installed apps, limiting permissions to the strict minimum and reading privacy policies are reasonable, accessible steps that don’t require a complete break with digital tools.

Further reading

See healthy information habits, setting up your phone for emergencies and the fictional scenario Montreal under cyberattack.