Skip to main content
0%

How AI Amplifies Surveillance of Users

7 min readUpdated

AI can automate searches for similar faces in a database and the analysis of user activity data. The results can inform identification, advertising, and other decisions. But a face match or a model's prediction can be wrong: it should not be treated as an established fact about a person. The following sections cover documented examples and ways to check data sharing.

From Manual Review to Mass Analysis

A face search across an image database is one example of automation. NIST distinguishes comparing two photos from searching for matches among many photos. In the latter case, a system can return candidates for further review.

To assess this process, distinguish:

  • Which images or other data have been collected
  • What the system compares or estimates
  • Who reviews the result and what action follows

Automated processing does not establish accuracy or justify collecting the data. Read more about centralized data storage risks.

Connection testHow fast is your internet right now?
Test your internet?Press to start — takes 30 seconds and up to 50 MB of data
—mspingmax—avg—min——Mbpsdownloadmax—avg—min——Mbpsuploadmax—avg—min—
Jitter—Ping under load—
Last speed test
MetricResultmaxavgmin
Ping—ms———
Download—Mbps———
Upload—Mbps———

Face and Identity Recognition

A database search can produce a false positive: the system links photos of different people. The opposite error is also possible: it fails to match photos of the same person. In NIST's 2019 study, results varied between algorithms and demographic groups. Those findings do not assess every system deployed today.

Identifying a person and inferring an emotion are different tasks. A 2019 review by Lisa Feldman Barrett and colleagues describes how interpretations of facial movements depend on context and vary between people and cultures. The same facial expression cannot automatically be equated with one emotion.

An image being in a database does not reveal its retention period, recipients, or permitted uses. Check those conditions with the operator of the particular system.

In December 2023, the FTC announced its case against Rite Aid. According to the Commission, its face recognition system generated false matches, after which employees followed customers, searched them, or ordered them to leave. This illustrates the consequences of an incorrect result; it does not establish equal accuracy across all systems.

Settings for protecting a personal device are covered in the smartphone security guide.

Behavioral Prediction and Profiling

A profile is the result of processing information available to a system, rather than a complete description of a person. Google, for example, explains using information and activity to personalize ads. Selecting an ad does not establish knowledge of the user's future actions.

In January 2025, the FTC published initial findings from its surveillance pricing study. The tools examined used location, browsing, and purchase information to tailor prices and offers. This describes the tools studied; it does not mean every store changes prices based on your profile.

Automated selection also needs scrutiny. In September 2023, the EEOC announced a settlement in the iTutorGroup case. It said software automatically rejected applicants based on age. The case illustrates the risk of a programmed selection rule; it does not by itself establish the use of a trained neural network. Not every automated decision is AI.

Data collection, profile building, and applying a model are separate stages. Read more about collection in the sections on browser fingerprinting and advertising profiles.

Who uses recognition and profiling

One organization can own a system while another uses it. In a GAO report published in June 2021, 20 of the 42 US federal agencies surveyed reported owning face recognition systems or using systems owned by others. The report also described the use of systems owned by non-government entities.

This is a documented example of interaction between organizations. It does not establish that every company's data is automatically available to the authorities.

For a particular service, check who collects information, who processes it, who receives it, and for what purpose. A shared sign-in across services does not reveal all these relationships. Read more about the risks of linking services to one account.

How to check permissions and data sharing

Start with specific data sources. App settings, ad personalization, and external cameras are controlled in different ways. Changing one setting does not establish that all tracking has stopped.

How to check an Android app's permissions

  1. Open Settings → Apps.
  2. Select the app. If it is not listed, open See all apps.
  3. Open Permissions.
  4. Select camera, microphone, or location access and check the allowed mode. If access is unnecessary, select Don't allow.
  5. Open the app again and check the features you need. If necessary, reconsider the permission for the particular feature.

The path and available modes can vary by device and Android version. Denying a permission controls that app's corresponding access; it does not delete records already shared or disable other organizations' cameras.

How to check Google ad personalization

  1. Open My Ad Center and check which account is signed in.
  2. Find Personalized ads.
  3. If you want to turn personalization off, select Off and confirm your choice.
  4. Open the setting again and check its state.

This is a Google ad personalization setting, not a switch for all analytics or AI. It does not stop every form of data processing or control other companies' advertising technology.

What a VPN changes

A protected VPN tunnel encrypts traffic between the device and the VPN server. For connections through it, the site sees the exit server's IP address. This does not delete an account profile, hide information sent to the site from the site itself, or stop face recognition by external cameras.

Additional measures are covered in internet security basics.

What to ask the system operator

  1. Find the data collection notice or service policy and a contact for privacy questions.
  2. Ask which data is used, for what task, how long it is retained, and who receives it.
  3. If a decision affected you, ask about available human review and correction of inaccurate information.
  4. Save the response and stated conditions. Check available procedures and rights for your country and circumstances.

An advertising claim that a system “uses AI” does not reveal its data sources or decision accuracy.

What to check after changing settings

Open the permissions list and selected ad setting again. Confirm that the intended state was saved and check that the app works. These checks show the state of specific settings, rather than the absence of all surveillance systems.

If a system's decision affected you, save the notice and request the available review procedure. Do not treat a prediction or face match as a proven fact about a person. Other measures are covered in the smartphone protection guide.

Author: — Cybersecurity specialist, VPNON