There is a quiet paradox at the heart of the modern digital economy. The same technologies that personalize your news feed, predict equipment failures in a factory, and recommend your next streaming binge are built on a continuous, largely invisible harvest of human behavior. You are the dataset, and the model recording never sleeps. This is not a conspiracy theory but the reality of AI’s engineering. Understanding exactly how AI systems collect, model, and monetize your digital life is no longer optional for practitioners, executives, or informed citizens who care about what happens to their data once it leaves their hands. 

The Rise of Digitalization and the Power Politics of AI 

The shift to digital infrastructure over the past two decades has not simply moved commerce and communication online. It has created an entirely new political economy, one in which data is both the raw material and the currency of power. Nations compete for AI supremacy. Corporations race to build the largest proprietary datasets. And the individuals who generate that data sit at the bottom of a very steep value chain. 

AI and machine learning models are, at their core, pattern-recognition engines trained on human-generated data. Every interaction you have with a digital system, whether you search for symptoms on a health portal, swipe through a dating app, or log into a corporate HR platform, contributes signals to models that infer far more than you explicitly disclosed. Your pace of scrolling, your hesitation before a purchase, the time you spend on a paragraph before closing the tab: all of it is a signal. Machine learning does not need you to answer a questionnaire. It reads the behavior instead. 

Governments have not been passive observers. The race for AI dominance between the United States, China, and the European Union has produced a geopolitical architecture in which data sovereignty is a national security matter. China’s data localisation laws, the EU’s General Data Protection Regulation, and ongoing US debates over federal privacy legislation all reflect the same underlying tension: who controls the data controls the future. Individuals caught in the middle often have the least say in the matter. 

India deserves particular attention. With 1.4 billion citizens, the world’s largest biometric identity programme in Aadhaar, and a government actively building domestic AI capability through its IndiaAI Mission, India is simultaneously the world’s largest democracy, navigating surveillance at scale and an emerging force in setting data governance norms for the Global South. The DPDPA grants data fiduciary obligations to organisations handling personal data, but critics note exemptions for government entities that leave citizen data exposed to state use without equivalent protections. How India resolves that tension will matter far beyond its borders. 

Edge AI and the Connected Device Environment: Surveillance Without the Cloud 

The most significant development in AI tracking over the past three years is not what has happened in data centers. It is what now happens at the edge: on your phone’s neural processing unit, inside your smart speaker, on the inference chip embedded in your car’s dashboard, and in the computer vision module watching the entrance of your favourite retailer. 

Wearables: Heart rate, sleep, stress, location, and activity patterns processed on-device and synced selectively to the manufacturer of cloud services. 

Smart Home: Voice assistants, thermostats, and security cameras build behavioral schedules of occupancy and daily routine. 

Connected Vehicles: Driving behavior, location history, in-cabin audio, and biometric seat sensors feed manufacturer data pipelines. 

Retail AI: Computer vision tracks dwell time, gaze direction, emotion proxies, and product interaction without any login requirement. 

On-Device LLMs: Small language models on phones process queries locally, but training telemetry and interaction metadata still flow upstream. 

Edge AI was sold to consumers on privacy improvement, offline functionality, and zero latency. Edge ML /AI models process directly on local devices rather than in centralized cloud data centers, and users assume data never leaves their premises. That framing is technically accurate and practically misleading. On-device inference reduces data transmission, but the models themselves were trained on cloud data; the outputs often sync to servers, and the metadata of what you do with an edge AI model is frequently harvested or monitored. As the inference stays local, the intelligence of your inference is not. Unlike web cookies and other trackers on the web, your prompts and data never leave your infrastructure. Edge AI did not move surveillance out of the clouds. It moved surveillance into every room by giving personal devices and people authority over their data. 

The connected device environment has created what researchers call ambient intelligence: an always-on layer of computation embedded in physical space that tracks behavior without requiring any active digital engagement from users. You do not need to be on your web browser for observation. You only need to have your device within your range.  

Cross-Device Identity Graphing: The Persistent Portrait 

Cross-device identity graphing links a single individual’s activity across multiple devices, platforms, and environments into one continuous, unified profile. It is the connective tissue of the modern tracking ecosystem, operating at a scale and sophistication that most users, and many practitioners, significantly underestimate. Identity graphs are built through two complementary techniques.  

• Deterministic matching leverages hard user identifiers: email addresses, phone numbers, and login credentials that definitively link activity across contexts. If you use your Gmail address to sign up for a retailer’s loyalty programme and that same address is your Google account login, your retail purchase history, search history, YouTube watch history, and location history are all linkable by any party holding that deterministic identifier. 

• Probabilistic matching is subtle, where inference happens with two devices likely belonging to the same person based on correlated signals: shared IP addresses at similar times, overlapping location patterns, common Wi-Fi networks, and complementary browsing behaviors at adjacent timestamps. The aggregate, processed through a machine learning model, assigns a confidence score to the hypothesis. Scores above a threshold are treated as confirmed identity matches. 

The result is that your desktop browser, mobile phone, tablet, smart TV, connected car, and wearable can all be woven into a single identity node in a data broker’s graph, without your knowledge, without your consent, and without any single identifier being shared across all of them. The graph simply infers the connection from the pattern of your life. 

Synthetic Inference via Generative AI: The Profile That Never Needed Your Data 

Generative AI has introduced a category of tracking risk that did not exist five years ago: synthetic inference. Large language models and multimodal generative systems can now construct detailed profiles of individuals not from data that those individuals directly generated, but from aggregated signals, statistical patterns, and the behavior of demographically similar populations. Every prompt typed into an LLM, every document uploaded to an AI productivity workspace, and every interaction with an autonomous software agent serves as telemetry. AI models do not just store information; they infer deep user traits. 

Synthetic inference also extends to the creation of entirely fabricated digital identities. Large language models can generate convincing online personas with plausible histories and writing styles indistinguishable from authentic human users, then deploy these at scale to populate platforms, influence discourse, and conduct social engineering. The same technology that writes your cover letter can manufacture a disinformation network. The role of ethics, genuine media, governance, and enterprise responsible AI (RAI) fits into the box. For practitioners building AI systems, the ethical weight of synthetic inference is significant. A model does not need to have been trained on an individual’s data to harm that individual through its outputs or commit fraud or legal offense with private data throughout the process loop. The obligation to audit for inference harms extends well beyond the training dataset. 

Data Monetization and Social Media Algorithm: Commercialization of Egalitarianism 

Data monetization operates through interlocking mechanisms most users never see. The most visible is targeting advertising: behavioral data used to serve ads statistically more likely to convert, meaning advertisers pay premium prices for specific audience segments, and the cost is incurred upon the end consumer. The economy runs far beyond advertising, with data brokers aggregating consumer records from thousands of sources, including purchase histories, public records, location data sold by mobile apps, and profile preferences from internet service providers. These profiles are sold to insurers assessing risk, employers running background checks, financial institutions making lending decisions, and political campaigns building voter models. The consumer whose data underpins these transactions rarely consents to the secondary uses, and in most jurisdictions has no right to compensation for the value created. 

Algorithmic platforms run continuous auctions across a global network of data brokers, applying identity graph merging to stitch together profiles from multiple sources before settlement. The exchange time is measured in milliseconds. The scale is measured in billions of identities transacted globally. The personalization promise, that targeted content serves your genuine interests, obscures a more complex reality. Optimization for conversion and engagement frequently produces filter bubbles, manipulative buying patterns, and demographic profiling that reinforces commercialization. When a platform learns that users in a certain area respond poorly to financial product advertisements, it may simply stop showing them opportunities, producing discrimination without any discriminatory intent encoded in the training data. Authenticity in personalization in user interest requires deliberate engineering choices that run against the grain of maximum revenue extraction. 

Tracking Inside the Enterprise: The Invisible Employee Dataset 

Some of the most consequential data collection happens inside organisations, and apathetic HR. Enterprise AI platforms and productivity suites now offer people analytics dashboards that measure engagement, predict attrition, score collaboration patterns, and flag anomalous behavior.

Enterprise AI models ingest email metadata, calendar density, document edit histories, meeting attendance, and response latency. None of this requires reading message content to derive sensitive inferences about performance, mental state, or career trajectory. In many jurisdictions, employees have no legal right to access, correct, or delete these profiles. The digital footprint you leave inside your employer’s systems may be more detailed than anything a consumer platform holds on you. 

Web Tracking Technologies: Far Beyond Cookies 

The deprecation of third-party cookies was widely reported as a privacy win. The tracking ecosystem responded by accelerating the adoption of techniques that are harder to block, detect, or regulate. 

Browser Fingerprinting 

Cross-Device ID Graphs 

Pixel Tracking 

CNAME Cloaking 

Server-side Tracking 

Privacy Sandbox APIs 

Browser fingerprinting combines dozens of device attributes, including GPU models, installed fonts, screen resolution, time zone, and plugin list, into a statistical identifier that tracks users across sessions without storing anything on their device. CNAME cloaking disguises third-party tracking scripts as first-party requests, bypassing browser cookie policies entirely. Google’s Privacy Sandbox APIs, intended as the cookie replacement, introduce audience cohort mechanisms that some researchers argue relocate the privacy problem rather than solve it. Server-side tracking moves data collection off the browser entirely, rendering client-side privacy tools largely irrelevant. 

Individual Data Sovereignty: Reclaiming Your Digital Self 

Individual data sovereignty refers to the principle that a person has the right to know what data is held about them, to control how it is used, to correct inaccuracies, to demand deletion, and ultimately to benefit from the economic value that their data generates rather than having that value extracted unilaterally. 

In practice, individual data sovereignty remains aspirational rather than operational in most parts of the world and commercial situations. Even where legal frameworks exist, consent interfaces are engineered to maximize agreement, deletion requests are honored slowly and incompletely, and the secondary markets into which data flows after initial collection are largely beyond the reach of individual redress mechanisms. The emerging response operates on multiple levels. Technically, privacy-enhancing technologies such as differential privacy, federated learning, and homomorphic encryption allow useful computation on data without exposing raw personal records. Legally, frameworks like GDPR, CCPA, and India’s DPDPA are progressively strengthening individual rights. Economically, data trusts and personal data stores, structures that allow individuals to pool and negotiate the terms of data use collectively, are gaining traction as a market mechanism for restoring some balance of power to individuals. 

For organisations, respecting individual data sovereignty is increasingly a business imperative rather than a regulatory burden. In highly sensitive industries like healthcare, banking, and insurance, the companies that will win defines AI skepticism, and privacy compliance is those that can credibly demonstrate that user data is handled with the same care they would want applied to their own. 

Critical Risks You Cannot Afford to Ignore 

• Algorithmic discrimination in insurance, credit, employment, and housing is derived from profiling rather than explicit demographic inputs 

• Re-identification attacks that deanonymise supposedly anonymised datasets using publicly available auxiliary information 

• Synthetic inference: GenAI models build accurate profiles of individuals from data that those individuals never directly provided 

• Cross-device identity graph exposure: a single data breach at a graph operator can expose the unified profiles of millions simultaneously 

• Edge device telemetry: on-device AI reduces cloud data transfer, but metadata about usage patterns continues to flow upstream 

• Inference attacks where models deduce sensitive attributes, including health conditions, sexual orientation, and political beliefs, from benign inputs 

• Synthetic identity fraud powered by LLMs generating convincing personas using real data fragments harvested from public sources 

• Cross-border data flows moving personal data into jurisdictions with weaker protections, beyond the reach of home-country law 

• Psychological manipulation through hyper-personalised content designed to exploit emotional vulnerabilities identified by behavioural models 

• AI-generated disinformation at scale: networks of synthetic personas capable of shifting public opinion 

Proactive Governance: Trust, Transparency, and the Accountability Gap 

Consent mechanisms are routinely designed to maximize opt-in rates rather than enable informed choice. Algorithmic impact assessments remain largely voluntary in most jurisdictions outside the European Union. Technology conglomerates and telemetry businesses have a responsibility that extends beyond legal compliance. Trust is built through proactive disclosures, transparency, and building guardrails to protect consumers: plain-language concrete data use policies, genuine opt-out mechanisms, especially when involving private businesses selling data to third parties for profits, user access to their own profile data, deletion that removes records from downstream markets, and algorithmic systems subjected to independent audit. The EU AI Act’s mandatory conformity assessments for high-risk AI systems are the leading example for governance that should eventually reach every major market. 

Regulatory risk, reputational exposure, and the erosion of consumer trust all impose costs that outweigh short-term revenue gains from maximalist data collection. Ethical data and AI frameworks in product development cycles, rather than bolting them on after regulatory pressure arrives, are now a strategic necessity.  

Protect Your Digital Sovereignty: 12 Actionable Steps 

• Use a privacy-focused browser such as Firefox with uBlock Origin or Brave, and enable DNS over HTTPS with a non-logging resolver such as Cloudflare 1.1.1.1 or NextDNS to prevent ISP-level query logging 

• Audit app permissions quarterly. Revoke location, microphone, contact, and camera access for any app without a clear functional need. Many apps request broad permissions speculatively at install and retain them indefinitely 

• For edge and connected devices, review your smart home hub’s data sharing settings and disable telemetry where the option exists. Segment IoT devices onto a separate Wi-Fi network to limit their access to your primary device environment 

• Use unique email aliases per service using tools such as SimpleLogin, Apple Hide My Email, or Fastmail masked addresses. This lets you identify which services sell your address and revoke access individually without exposing your primary identity 

• Request your data from major data brokers and submit opt-out requests under GDPR, CCPA, or DPDPA where applicable. Automated services such as DeleteMe can streamline this process in supported markets 

• Disable ad personalisation at the platform level across Google My Activity, Meta Ad Preferences, Apple Personalised Ads, and Amazon advertising settings. These toggles are meaningful and reduce the richness of profiles built on your behaviors 

• Be deliberate about cross-device login behavior. Using the same social login across unrelated services creates explicit deterministic links in identity graphs. Maintain separate accounts or use temporary login tools where possible 

• Treat your connected vehicle as a data source. Review your manufacturer’s data sharing consent settings and understand what is collected by default, including location history, driving patterns, and in-cabin data, before it flows to third parties 

• Apply data minimalism as a decision filter: before signing up for a service or installing an app, assess what data it collects, how it is monetised, and whether the value exchange is genuinely fair to you 

• Use a VPN from a verified no-log provider on public networks, while understanding that a VPN shifts trust to the VPN provider rather than eliminating exposure. It is one layer of protection, not a complete solution 

• For enterprise AI practitioners: advocate for privacy by design. Data minimization, differential privacy, federated learning, and on-device inference are engineering choices available today. Resist organizational pressure to treat user data as a free resource 

• Stay informed about your legal rights under applicable frameworks. GDPR’s right of access, rectification, erasure, and data portability, India’s DPDPA data principal rights, and CCPA’s right to know and opt out are tools available to you. Use them proactively, not reactively 

Our Conclusion Clincher 

The tracking ecosystem is not a malfunction in the modern economy. It is a digital design choice, made every day by engineers, product managers, and investors for the economy and its people. The emergence of Edge AI, synthetic inference via generative models, and cross-device identity graphing has fundamentally changed the scale and intimacy of digital surveillance now possible. Tracking no longer requires a login, a cookie, or even an internet-connected action. It requires only proximity to a device that has been instrumented from end to end.

The geopolitical contest over data governance, now explicitly involving the United States, China, the European Union, and a rapidly asserting India, will shape whether individual data sovereignty becomes a meaningful right or remains an aspirational phrase in regulatory preambles. Its outcome will be determined in part by whether practitioners, executives, and ordinary citizens demand accountability from the systems they build, fund, and use.

Your digital footprint is valuable, and it is meaningful to start with knowing the new complicated reality and assertively gaining technical literacy, legal awareness, and deliberate decision-making in our daily lives.

<p>The post You Are Being Tracked Almost Everywhere: How AI Keeps Score on You first appeared on Tech AI Magazine - The World's Leading AI Magazine.</p>