The systems that learned what show you would watch next also learned what makes you angry, when your judgment is most likely to slip, who you trust without verification, and what kind of message you are least likely to question. That data was collected for commercial purposes. It is increasingly used for considerably more than that.

This series examines the mechanism, the harm patterns it produces, and what it means for the people operating in an environment where the infrastructure of personalization and the infrastructure of influence have become the same thing.

6
Chapters examining personalization from behavioral data to adversarial use
3 sec
Audio sample required for a credible voice clone of a known individual
70%+
Accuracy with which behavioral data can predict psychological attributes without disclosure

“The question is not whether personalization works. The question is who it ultimately serves, and under what conditions the answer to that question is the person whose behavior is being modeled rather than everyone else with access to that model.”

Kia Hakimi — Chapter 6

Chapter 1 From Convenience to Control

Personalization shifted from reactive recommendation to active behavioral steering before most people noticed the transition. The systems that surface content based on what you have done are now predicting what you are likely to do next and positioning prompts, offers, and content to increase the probability you do it. The infrastructure of convenience and the infrastructure of influence are no longer separate things.

The same infrastructure built to help you find a good show is also capable of mapping your psychological vulnerabilities and exploiting them in real time.
Read Chapter 1 →
Chapter 2 What They Actually Know About You

The data layer goes considerably deeper than most people assume, and the more important question is not what has been collected but what has been inferred. Behavioral signals including clickstreams, dwell time, scroll depth, device fingerprints, and geolocation patterns support inferences about political orientation, emotional state, income, relationship status, health conditions, and psychological traits. The shift from demographic targeting to psychological modeling represents a qualitative change in what personalization is capable of doing.

The system does not need to be right about you specifically. It needs to produce better outcomes than chance across large populations, and by that standard these systems are working very well.
Read Chapter 2 →
Chapter 3 The Harm Patterns

The harm patterns that emerge from weaponized personalization are not random. They follow from what gets collected, what gets inferred, and what systems are optimized to produce. Compulsion loops engineered to specific psychological profiles. Financial manipulation timed to moments of reduced resistance. Predatory targeting of people flagged as vulnerable. A polarization gradient that no one designed but that engagement optimization reliably produces.

Most people notice when personalization crosses into visibility. The more significant influence happens below that threshold, shaping what you see, consider, and decide without producing any signal that something is occurring.
Read Chapter 3 →
Chapter 4 Trust as the Target

When the objective shifts from attention or money to trust itself, the nature of the harm changes. Behavioral profiling produces detailed maps of trust relationships: who gets an immediate response, whose requests move through without friction, whose communications produce action rather than deliberation. Account compromise allows an adversary to inherit the accumulated trust an account has built with its entire network. Synthetic impersonation at scale is no longer a future risk for high-visibility individuals. The data required to construct it has largely already been assembled.

The highest-value trust relationships from an adversarial standpoint are not the most prominent ones. They are the closest, most habitual, and least subject to verification.
Read Chapter 4 →
Chapter 5 Why Good Intentions Are Not Enough

Most personalization harm is not the product of malicious intent. It emerges from optimization systems discovering the most efficient path to a metric, distributed organizational responsibility that makes aggregate harm invisible at the level of any individual decision, and incentive structures that are central to the economic model rather than incidental to it. The track record of voluntary internal reform is limited. The changes that have produced meaningful movement have generally come from external pressure, not internal goodwill.

Intent does not constrain what optimization discovers. The system learns what works on you without anyone deciding to manipulate you specifically.
Read Chapter 5 →
Chapter 6 What Comes Next

Agentic AI systems are shifting personalization from reactive surfacing to autonomous action on behalf of users, with a data layer significantly richer than anything passive browsing produces. Behavioral inference is expanding beyond digital into insurance, credit, employment, and healthcare. The regulatory directions with the most promise address structural problems rather than surface symptoms. For individuals, the more important intervention is at the level of understanding exposure rather than adjusting settings.

The gap between the capability of personalization infrastructure and the governance of that capability is widening rather than narrowing. Understanding that gap is where a grounded response has to start.
Read Chapter 6 →

Shadow Sciences Group

Understanding Your Exposure in This Environment

For high-visibility individuals, the behavioral record that personalization systems have assembled is not a consumer profile. It is an exposure map. The Strategic Exposure Assessment is designed to evaluate what that map looks like from the outside and where the gaps between self-perception and behavioral footprint are most consequential. Confidential introductory consultations are available.