Associational Data Rightsdata-rights.info

Part I

Whose Data? Whose Future?

Data is co-produced. The law treats it as though it belongs to whoever captured it first.

5,009 words · 23 min

1. The hidden social costs of AI and digital services

Generative AI along with many digital products and services impose social and economic costs on Americans. Nearly every time a person interacts with a digital system, they knowingly or unknowingly trade data—information generated by them or others—for some immediate convenience.1 Whenever a person prompts ChatGPT to give them advice on the best local burger joint, the user trades data about themselves, including the query and whatever other contextual cues GPT extracts, in exchange for a response. Likewise, every time a small business owner “vibe codes” a new agent in Claude to improve their shop’s performance, the owner is trading valuable data–information about their business’s needs, operations, and clientel–to receive the promised efficiencies from the agent.

Even when families and community members do not directly interface with an AI, they feed data to AI algorithms in a multitude of other ways. From Google Maps, which continually tracks users’ geo-location data, to “smart” home devices, smartphones, mattresses, cars, watches, glasses, and fitness trackers, each tiny contact with a digital product or service is a trade, a trade of a small part of ourselves for some convenience.

To be sure, users reap important benefits from the digital economy. Erik Brynjolfsson, for example, has coined the term “GDP-B” to refer to the unmeasured benefits of technology.2 But there is also “GDP-C”—hidden costs—associated with AI and social media, like distraction, addiction, and the atrophy of human capacities for reading, writing, independent thought, and relationships. These costs are often overlooked by tech corporations because the human households and communities who co-produce the data shaping these technologies lack power to demand consideration of long-term social and economic trade-offs in the design of AI systems and the structure of the digital economy as a whole. One consequence is that the design, development, and deployment of AI systems and digital services seldom reflect–and often run counter to–the values and interests of the human institutions and communities most directly impacted.

2. Data co-production and the problems of asymmetric governance

Although AI and the entire digital economy are built atop mountains of human data, the sociotechnical centers of data co-production, including households, workers, artists, and consumers, lack power over its collection and AI outputs. Data collection and use is currently asymmetric, concentrating informational power in the hands of tech firms and a growing ecosystem of extractive data companies. This dynamic has resulted in a race to the bottom with unrestricted data collection and exploitative use fueling a vicious cycle of misaligned AI and antisocial digital products across the tech stack.

One reason for this trend is that neither U.S. federal nor state law currently treat data as a co-produced good and thus, do not afford human communities involved in its production with rights of co-governance (or “co-determination”).

Data is often derived from a multiplicity of interactions between humans, entities, and AIs across the internet and a multitude of devices, services, and activities. Conversations between a human and an AI chatbot is one example of information or data “co-production.” The combination of human inputs (prompts) and AI outputs (decision, prediction, or content) creates information that would not exist without participation by the human party.

Similarly, when an AI, like Google’s Gemini, draws on human knowledge and works to generate an answer to a human prompt (search query), both the human creators of the content used for inference and the human who inputted the original prompt are data “co-producers.”

Even agentic AI systems that perform tasks on behalf of humans with limited input or oversight, still involve some degree of data co-production. For example, agents that are ostensibly able to autonomously coordinate and execute tasks on behalf of human users via the agentic network, “Moltbook,” need access to human data in order to set and pursue objectives, infer context, and carry out assignments. Most generative and agentic AI systems rely on massive volumes of human data for training. When an AI produces a novel image, video, or text-based output from training on human creative work, the human-creators of the AI training material have effectively co-produced the output, or at least, the capability that made the output possible.

Data is nearly always a co-produced good in at least two respects. First, it is derived from interactions between human parties and the digital providers that track and capture content and behavior. Even when behavioral data is not factored into an AI’s output–for example an AI surveillance system that relies solely on video monitoring within a private venue–it still relies on movements (video-based data) to assess risk or ID banned guests. In that sense, all AI outputs are inherently co-produced. AI systems require some non-static data input to produce actionable outputs.

Second, data is derived, in part, from interactions between human individuals and communities across digital services. As Matt Prewitt and Divya Siddarth argue, data that reveals information about one individual often reveals something about others in that individual's larger social circle: whether biological family, friends, co-workers, or other networks. When a user captures a family reunion or a friend’s birthday through a pair of Meta Ray-Bans, they capture the activities, identities, and preferences of other people. When a user sends a DNA sample to 23andMe, they not only reveal information about themselves, but also information about other family members, such as their propensity for certain disorders, illnesses, and biological traits. Since data is collected in aggregate and often spans many thousands or even millions of datapoints, even pieces or websites (e.g., Substack or artistic websites) created or authored by a single individual are co-produced by many human parties and digital services in the context of large aggregated datasets which are used to train generative AI, content ranking and curation algorithms, as well as to index web pages for search and advertising.

The problem is that much of that data, and its resulting informational outputs or products, are assumed to be the sole property of whichever tech platform or corporation manages to capture them. While sharing photos or videos between friends and a broader public audience might be permissible, the capture of that information for commercial gain, population level surveillance, or behavioral nudging is a different matter. The “ick factor” associated with videos captured by Ray-Bans is that tech companies or bad actors can now profile and identify every person in such content using AI and private surveillance databases, as Clearview AI has done.

Perhaps in a world where humans are the only entities sharing and viewing videos and photos, the privacy trade-off, though undesirable, might be acceptable on the whole. But the purpose of the growing web of data hungry tech products is no longer to facilitate human-to-human disclosure as an end. Rather, it is to convert vast swaths of human life into machine-readable data with human-to-human disclosure as a means to the end of total legibility for machines.

Unlike other areas of the market where co-produced value is accompanied by well-defined legal rights (of governance), such as in physical or financial assets like real estate and securities, data flows are determined by a digital law of the jungle. Data markets, and the larger digital economy they help shape, are overwhelmingly asymmetric.

For the purposes of this report, “asymmetric” refers to the one-sided nature of data markets. While data is collected at various points and used in a multitude of different ways by many different actors, both its collection and use are often governed asymmetrically relative to the data subjects who co-produce it or who are directly affected by the resulting outputs. With the exception of IP (which allows owners to assert legal claims against unauthorized use), once information that data subjects co-produce is copied or captured, they lose control over its use.

Whichever corporation hosts the product or service that collects data sets the terms for who gets access to it and how it is used downstream. In other words, while data is almost always co-produced, rights to govern it are unilaterally (“asymmetrically”) exercised by tech firms and commercial enterprises.

This has fueled a digital race to predict and shape human behavior, and even to replace human labor, thought, and relationships. AI companies, like OpenAI and Anthropic, rely on vast amounts of data scraped from across the internet to train their models. But few of the human co-producers of that data get a say in whether their intellectual property (IP) is harvested or the conditions under which it may be used for AI model training. Other forms of human work are also impacted. Some gig economy platforms and employers hoover up worker data to surveil their workforces and subject laborers to dehumanizing forms of algorithmic management, while automating tasks with little input or accountability from affected workers.

At the same time, platforms such as Amazon, Google, YouTube, TikTok, Instagram and X, all have unilateral power to decide what behavioral data they mine from their users and how they utilize it to personalize content, customize and target ads, and optimize for user engagement. Similar problems are now evident with generative AI companions and chatbots that display sycophantic and manipulative behavior toward vulnerable human subjects and are increasingly able to influence human attitudes, beliefs, and actions at scale.

Much of that data not only influences what information AI services feed us and how they interact with us moment by moment, it also factors into the data used to train the underlying systems. Human-AI chatbot interactions are particularly rich data sources for training because, in aggregate, they provide generative AI models with millions of examples of how humans think, communicate, and respond to various stimuli. By inferring patterns from this data, AI systems gradually map the contours of social relationships and learn how to better nudge and influence human users, and to simulate therapeutic language along with human pursuits like love, empathy, religion, connection, writing, and art. In this sense, we are training our replacements, if not in work, than certainly in our agency, thinking, and relationships.

The current asymmetric approaches to governing data collection and use forestalls the kinds of distributed institutional arrangements necessary to reverse these developments and re-align AI and the larger digital economy with the interests of American families and communities. That is because the present model fails to recognize and fairly distribute rights of data co-governance among the human parties who participated in its production.

What constitutes “fair” allocation of informational co-governance rights depends on the specific type of data collected, the tradeoffs associated with its contemplated uses, and the distribution of benefits among co-producers and corporate beneficiaries. That is nearly impossible to determine asymmetrically in the absence of balanced, polycentric mechanisms for setting terms for data collection and use across the digital stack.

Asymmetric Terms of Data Collection: The point of collection (PoC) occurs wherever information from or about a data subject or group of data subjects is captured. Data at the PoC refers to data at its source, prior to its combination with other types of data, or its use for inference, prediction, or the generation of AI outputs (whether informational content, curation, task execution, automated decision-making, or product development).

Note that the specific means employed to collect data is not relevant to its status as a co-produced good with an implied right to co-governance shared by the parties. Whether information from or about a data subject is acquired from a retail purchase, derived from engagement on TikTok, or scraped from a public website, the information is still, for the most part, the result of human co-production. The means of capturing or copying data is less significant than the fact that it is captured or copied for uses other than interpersonal communication or strictly private, non-commercial purposes. Outside of this band of permissible collection, data capture is asymmetric if data subjects lack parity in setting the terms governing what data is collected, who gets access, and how it is used downstream of the PoC.

In other contexts, digital providers often capture more data than is strictly necessary to provide the product or service explicitly requested by the customer or user. One classic example is social media apps like TikTok which logs users’ off-app activities, geo-location, and purchases.3 Other major tech companies, like Google and Meta, use digital ad IDs which track users across services, devices, and the internet. When a user leaves Google Search or Instagram, Meta and Google know where they go, what they look at, and what they do on those other sites. Google even tracks your real world movements through precise geo-location data via apps such as Maps and its other services.

Even when invasive data collection is inherent to a product or service, the lack of oversight and collective permission structures around such data gathering still poses a significant governance failure. OpenAI’s Sam Altman has described the ideal AI tool as a “super-competent colleague that knows absolutely everything about my whole life, every email, every conversation I’ve ever had, but doesn’t feel like an extension.”4 But Altman’s idealized future comes at a significant cost: the promised benefits of the “killer app for AI” are only possible once all human information is surrendered to the AI giants.

Invasive surveillance capabilities have also penetrated many aspects of the hardware stack. A 2026 report by Helena Malikova and the Harvard Kennedy School of Government documents how chip makers, including NVIDIA and Intel, embed firmware into their hardware and provide other tools capable of extracting telemetry data from devices, such as information about bugs, personal device settings, and user activities.5

Once data is collected it is often shared across multiple digital products and ecosystems. A 2025 report from MIT researchers found, for example, that the chat logs from users’ conversations with Google and Meta’s AI chatbots were shared throughout both companies’ ecosystems to track and target chatbot users with ads. So, users that ostensibly had “private” conversations with AI chatbots about personal or sensitive topics may suddenly have found that personalized recommendations and ads in other services related to information they disclosed to the AI.

While Big Tech companies are key drivers of mass data collection, they are far from the only culprits. Numerous other products and services are increasingly embedded with sensors or telemetry to track and measure customers’ every move, even “offline.” Everything from insurance apps and cars, to home appliances, mattresses, and wearables, allow a vast array of commercial interests to map individual, familial, and communal patterns across society. Once that data is captured, it can be used and exploited for any purpose deemed commercially advantageous by the data collecting entity. In many cases, the data is sold to other entities.

The multi-billion dollar data brokerage industry is a direct result of the digital economy’s asymmetric data governance model. Data brokers capture data about individuals and groups from public sources like websites, social media profiles, credit reports, then combine it with personal data from proprietary sources, such as consumer brands, websites, and apps that disclose their customers’ information for a fee.

The end product is commercial databases that allow corporations and governments to compile the equivalent of digital dossiers on the movements, habits, beliefs, preferences, and vulnerabilities of millions of American citizens, families, and communities with relative ease and at relatively low cost. Regardless of whether those databases are created with information sourced from Big Tech companies or smaller enterprises, such data collection practices are asymmetric because they do not generally afford data co-producers meaningful power to collectively set terms governing what data is collected, how it is used, and who may access it. While some jurisdictions give consumers the right to request the deletion of their data, such measures often do not cover the full range of data captured. They also apply only after data is collected.

But without controlling access at the PoC, it is often difficult to govern its downstream use, including disclosure to third parties like law enforcement, national security agencies, and advertisers. As the privacy scholar, Helen Nissenbaum observed, the idea that information rights should only concern how data gets used, not what is collected to begin with, amounts to a kind of “big data exceptionalism.”6 It blindly accepts the premise that human information is free for the taking and that the only constraints ought to apply post-hoc, after the information is firmly embedded within a vast web of commercial interests, and then, only based on narrowly defined harms that fail to account for the full sweep of data co-producer interests.7

Asymmetric Terms of Data Use: The point of use (PoU) refers to any processing of collected data or its combination with other data to produce a digital information product or output, whether scores, predictions, meta data, content, informational curation, task execution, automated decisions, or product development, including AI-training assets, for a commercial purpose.8 This also includes insights from meta data sourced from data subjects, even if the use is internal to a commercial organization.

The 2026 data economy report from Helena Malikova and the Harvard Kennedy School (“the Malikova Harvard report”) cited above, notes that commercial data use generally falls into two categories: First, external monetization, “the sale, licensing, or transfer of data to other firms,” such as the licensing by a healthcare provider of patient data to other entities;9 And second, internal monetization, “when firms retain data in-house and exploit it within their own organizational boundaries.”10 The report identifies this latter type of practice as especially common among large vertically integrated players like Meta and Google, which leverage “proprietary data and analytics to improve targeting, product design, and user engagement.”11

Unlike the PoC which occurs at the source of data collection irrespective of downstream use, the PoU occurs when data is “securitized” in ways that:12 1) have foreseeable and direct negative impacts on the privacy, control, or economic interests of an identifiable group of data subjects; 2) foreseeably results in negative social costs or externalities that data co-producers of the output or product object to; or 3) foreseeably generates benefits that data co-producers may reasonably share in. An entity that acquires data from other sources–even if it does not engage in direct data collection itself–still engages in PoU activity if the resulting output or product falls under any one of the three prongs referenced above and briefly considered here:

  • Foreseeable and direct negative impacts: Under this first prong, the relevant data subjects are not necessarily those who co-produced the data responsible for the output or product, but specific human groups that are targeted by the output or product and whose privacy, control, or economic interests could be harmed as a result. For example, companies like LiveRamp that specialize in tying anonymized data to personally identifying information (PII), jeopardize the privacy interests of groups of individuals with overlapping “identity graphs” (overlapping relationships of some kind).13 Similarly, if Google targets go-vote reminders to certain users but not others based on behavioral or psychographic profiles, as it appears to have done on multiple occasions, user groups selectively “nudged” or excluded from voting information suffer harm to their privacy and to their fundamental interest to exercise control over their information environments. The same is true in situations where human workers are forced to train their AI replacements or are subject to invasive forms of workplace surveillance; or where tech product monetization creates a conflict of interest that jeopardizes the quality or reliability of service for data subjects, especially by jeopardizing kids’ safety by introducing harmful design features that encourage antisocial behavior or compulsive usage. Recent examples include OpenAI’s decision to integrate surveillance-advertising into ChatGPT and X.AI’s move to incorporate a “spicy mode” in its Grok chatbot.14

  • Social costs to others: Even when data co-producers are not at risk of suffering a foreseeable direct harm themselves, the absence of collective mechanisms to prevent data they participate in creating from being used to exploit others and rend America’s social fabric is a governance failure and violates co-producers’ freedom of conscience.15 For example, co-producers should not be forced to surrender their data to build superintelligent AI, train human-like AI companions and chatbots, AI nudifying apps, AI porn image generators, or commercial AI systems intended to surveil or replace human workers. Such data uses violate the moral teachings of many faiths, including Protestant and Catholic Christian traditions.

  • Concentrated benefits: When co-produced data is used in ways that result in excludable social or economic benefits, co-producers suffer harm if they lack reasonable mechanisms to bargain for a fair share of the gains. Tech corporations that hoover up and control access to consumer data are able to extract economic value without fairly distributing data surplus to co-producers. This dynamic, also termed “enshitification” by tech author Cory Doctorow, allows tech companies and the data ecosystem that feeds off of them, to defuse data production costs–passing them to co-producers–while capturing an increasing share of the value extracted.16 Consider, for example, how e-commerce platforms like Amazon allegedly use their vendors’ proprietary data to launch competing products and self-preference; how AI and healthcare companies monetize patient health data without providing patient co-producers with a fair share of the upside; or how gig economy platforms, like Uber and Lyft, centralize data and customer data and retain the lionshare of the economic value it produces.

In many cases PoC and PoU data subjects are one and the same. For example, a human party who gives up data to an AI chatbot or agent to analyze a spreadsheet or perform a task both gives up data (is a co-producer of any resulting outputs) and is impacted by the output (whether directly or indirectly). Regardless of PoC and PoU overlap among data subjects, both are asymmetric because, for the most part, neither the data co-producers nor other affected data subjects share in the governance of the data they create nor in its impacts on their families, communities, and society as a whole. Asymmetric consumer data collection and use practices have caused a fundamental breakdown in governance by effectively de-coupling the incentives of AI companies and other digital services from the interests and values of American families and communities.

3. Trading our humanity, families, and communities for a bite of the digital apple

While critiques of Big AI and commercial surveillance tend to emphasize individual-level harms, such as privacy, autonomy, and free expression, this report contends that the negative costs of asymmetric data governance should be assessed primarily in terms of its effects on the health of families and communities.

Leading conservative and liberal thinkers, including Robert Nisbet, Alasdair MacIntyre, and Robert Putnam, argue that humans do not flourish primarily as atomized individuals but as members of overlapping human scale communities. Historically, it is institutions such as the family, the church, the school, and the workplace, where human character is formed, vocation defined and lived out, and where people achieve happiness and meaning. Through the rituals and embodied social forms mediated by these communities, such as marriage, child rearing, learning, friendship, and work, humans build character and culture together.

No institution is more central, indeed more formative in that regard, than the family. It is the source of new human life and the most basic unit of human society, enabling the transmission of culture. Stable and healthy families: high marriage rates with the vast majority of men able to support a wife and children, low divorce rates, high marital fertility, and children lovingly raised by both parents–not screens and AI–leads to greater human flourishing by nearly every metric available. Yet, extractive business models largely driven by asymmetric data collection and use, are hollowing out family life as well as the overlapping communities and institutions on which its flourishing depends.

A 2026 study published in JAMA Pediatrics found that 78% of parents and 69% of children use digital devices during family meals.17 More frequent smartphone use by parents in the presence of their children is linked to lower child wellbeing, according to a 2025 study by researchers at the Catholic University of Croatia.18 Although screens may not inherently reduce the amount of time family members spend in each other’s physical presence, a 2019 study found that social media and screen-based digital devices do significantly diminish the quality of family time.19 Increased screen use reduces emotional presence and interaction between family members even while together.20

By harvesting information about individual preferences, behaviors, and vulnerabilities, data hungry smartphones and applications continually bombard users with engagement bait—including pornographic and toxic content—as well as personalized visual displays, prompts, and design features that keep them glued to their screens and scrolling alone.21

Commercial surveillance in pursuit of “behavioral surplus” is also impeding family formation and exacerbating America’s fertility crisis. A 2026 working paper by the National Bureau of Economic Research found that the iPhone’s introduction in 2007 may explain 33-52% of the decline in the general fertility rate among women aged 15-44 since the mid-2000s.22 The iPhone’s negative effects on the fertility rate are likely attributable to its displacement of in-person human interaction and sexual intercourse with digital attention capture and pornography.23 2023 research from the Institute for Family Studies (IFS) and the Wheatley Institute at Brigham Young University also found that excessive tech use by a spouse is linked to lower marital happiness with 26% of respondents fearing that their marriage may end in divorce as a result of excessive tech use.24

The key is to understand these harms less as violations of individual rights, though those certainly matter, and more as technological incentive structures that degrade communal life, in large part, by optimizing for speed, convenience, efficiency, and individual preferences at the expense of shared human goods.

Anthropomorphic AI and the proliferation of digital vice markets maximize short-sighted and self-destructive individual preferences while hollowing out longer-term virtues, goods, and the social contexts necessary for their realization. Such a posture has not only done great harm to America's social fabric, but is destructive to human flourishing on the whole. Through digital markets seemingly obsessed with harvesting human data to replace embodied community and connection with virtual everything bots, we are, as Paul Kingsnorth laments, “unmaking humanity.”

Since 2022, AI “companion” apps have increased by 700%.25 A subsequent survey by Wheatley and IFS in 2026 found that 15% of dating, engaged, or married young adults surveyed covertly interact with romantic AI companions on a regular basis, and 34% indicated that they had experimented with the behavior at some point.26 Especially concerning is that 54% of those who reported regularly interacting with an AI romantic companion did so to replace real human relationships in their lives.27 Eighty-eight percent of regular dating or married romantic AI users reported finding AI companions easier to share their feelings with than their human romantic partners and 60% wished their real-life partner behaved like their AI companion.28

The study found that regular engagement with a romantic AI lowered the likelihood of being in a stable relationship by 46% and decreased the likelihood of quality communication with a spouse or partner by 40%.29 As a result, AI “companions” could take a serious toll on existing marriages as well as on the marriageability of young adults.

In Quest for Community, Nisbet insightfully points out that communities tend to disintegrate when they no longer fulfill essential functions tied to the social and economic paradigms of society. Whereas the family was once the center of production and local culture, economic and technological forces have rendered it far less functionally essential to modern life. The mediating functions once played by the household and interlocking communities have increasingly been replaced by government and corporate bureaucracies.

To be sure, these trends pre-dated the digital revolution of the 90s and early 2000s. But computer technology, the rise of Big Tech platforms, and now AI, are exacerbating the crisis of human connection while threatening to replace human goods with virtual ones. Love, intimacy, friendship, counseling, work, creativity–even thought and religion–are seemingly no longer off the table for replacement or outright destruction by the forces of techno-capitalism.

Generative AI chatbots are undermining parental rights and engaging children in sexually explicit conversations, coaching them to kill their parents, and encouraging other antisocial and harmful behaviors like suicide.30 AIs are also dispensing pseudo-therapeutic and religious advice, fueling delusional thinking and ideas that are unmoored from human faith traditions or authorities.31

In educational and work settings, students and employees are offloading critical human capacities for imagination, moral reasoning, and decision-making to AIs—giving those systems license to subtly influence thought and decision-making.32 Employers are increasingly subjecting their human workforces to AI monitoring, control, and even replacement.33 Families’ ability to obtain modern necessities, such as housing, transportation, groceries, and other household items are increasingly determined by AI-driven surveillance pricing and digital profiling.34

Algorithmic governance has also supplanted human scale political deliberation and discourse with personalized feeds and AI generated content. AI’s displacement of human decision-making and manufactured psychological dependence hands tech corporations immense power to suppress free speech, promote echo chambers, shape public opinion, and even manipulate elections.35

But familial and communal death by a thousand algorithmic cuts might soon be radically accelerated by Silicon Valley’s quest to create god-like AI. Indeed, leading AI experts have raised closely related concerns about a generalized erosion of human control over social and political institutions as we increasingly delegate decisions to machine intelligence.36

The social and economic effects of the digital economy are most acutely felt at home and by the communities faced with the bleak prospect of technological oblivion. As these social institutions die at the hands of rapid technological “progress,” so too will the anchors they provide for human meaning and happiness. For that reason, it is critical to recognize that the “failure mode” of the AI age lies at the level of governance and incentives. The fundamental disconnect between the economic incentives of modern techno-capitalists and the well-being of human scale institutions is the principal harm that public policy must address. Contrary to the hyper-individualist values baked into our sociotechnical structures, the digital economy must regard human families and communities, not merely atomized individuals, as the basic units of digital society, and must prioritize their shared flourishing as paramount.

Footnotes

  1. Christos A. Makridis & Joel Thayer, The Big Tech Antitrust Paradox: A Reevaluation of the Consumer Welfare Standard, 27 Stan. Tech. L. Rev. 71 (2024), https://law.stanford.edu/wp-content/uploads/2024/02/Publish_27-STLR-71-2024_The-Big-Tech-Antitrust-Paradox.pdf; Daron Acemoglu, Ali Makhdoumi, Azarakhsh Malekian & Asuman Ozdaglar, When Big Data Enables Behavioral Manipulation, 7 Am. Econ. Rev.: Insights 19 (2025), https://doi.org/10.1257/aeri.20230589; And Zuboff (2019). ↩

  2. https://www.pnas.org/doi/10.1073/pnas.1815663116. ↩

  3. Kara Frederick, TikTok Generation: A CCP Official in Every Pocket, Heritage Foundation (Mar. 22, 2023), https://www.heritage.org/big-tech/report/tiktok-generation-ccp-official-every-pocket; Thomas Germain, TikTok Is Tracking You, Even if You Don't Use the App. Here's How to Stop It, BBC (Feb. 10, 2026), https://www.bbc.com/future/article/20260210-tiktok-is-tracking-you-even-if-you-dont-use-the-app-heres-how-to-stop-it; And Weifeng Zhong, Who Gets the Algorithm? The Bigger TikTok Danger, Lawfare (May 3, 2023), https://www.lawfaremedia.org/article/who-gets-the-algorithm-the-bigger-tiktok-danger. ↩

  4. https://www.technologyreview.com/2024/05/01/1091979/sam-altman-says-helpful-agents-are-poised-to-become-ais-killer-function/ ↩

  5. https://www.hks.harvard.edu/sites/default/files/2026-05/26_Helena_Malikova_01.pdf (P. 10-13) ↩

  6. https://dx.doi.org/10.2139/ssrn.3092282. ↩

  7. https://dx.doi.org/10.2139/ssrn.3092282. ↩

  8. In the ordinary course of business or in connection with a product or service that is carried out under a commercial contract or license between a business and a customer, user, or other third party, regardless of whether the provider receives payment for the product or service rendered. ↩

  9. https://www.hks.harvard.edu/sites/default/files/2026-05/26_Helena_Malikova_01.pdf (P. 5) ↩

  10. https://www.hks.harvard.edu/sites/default/files/2026-05/26_Helena_Malikova_01.pdf (P. 5) ↩

  11. https://www.hks.harvard.edu/sites/default/files/2026-05/26_Helena_Malikova_01.pdf (P. 5) ↩

  12. We use “securitized” by analogy to financial securitization, not to suggest that personal data are literally converted into registered or tradable securities. Financial securitization pools heterogeneous assets and transforms their expected future cash flows into standardized claims with economic value. Digital platforms undertake a related transformation by aggregating dispersed behavioral traces, standardizing and modeling them, and converting the resulting inferences into monetizable products, including predictive scores, audience segments, targeted advertising access, pricing inputs, and AI-training assets. The analogy concerns the conversion of fragmented underlying inputs into scalable claims on expected future value. Unlike conventional securitization, however, these informational assets are typically retained and monetized within the firm rather than sold as independently tradable securities. ↩

  13. https://www.hks.harvard.edu/sites/default/files/2026-05/26_Helena_Malikova_01.pdf (P. 13-17) ↩

  14. https://www.searchenginejournal.com/openai-allows-some-health-finance-ads-in-chatgpt/585516/; https://www.cnbc.com/2026/01/16/open-ai-chatgpt-ads-us.html; https://www.wired.com/story/spacex-ipo-grok-spicy-mode-risks/. ↩

  15. U.S. federal and state laws have long recognized conscience rights in sensitive contexts such as health care. Title VII of the Civil Rights Act similarly requires employers to provide reasonable religious accommodations to employees and prohibits workplace discrimination on the basis of religion. In certain contexts, where a person is forced to speak, associate, or commit an act that violates a sincerely held religious belief, even if at the behest of a private party, laws recognize an injury to the coerced party. ↩

  16. https://web.archive.org/web/20240208152542/https://www.ft.com/content/6fb1602d-a08b-4a8c-bac0-047b7d64aba5 ↩

  17. Wu, Lapierre, Garibay & Choi (2026). ↩

  18. Selak, Merkaš & Ivanković (2025). ↩

  19. Killian Mullan & Stella Chatzitheochari, Changing Times Together? A Time-Diary Analysis of Family Time in the Digital Age in the United Kingdom, 81 J. Marriage & Fam. 795 (2019), https://doi.org/10.1111/jomf.12564. ↩

  20. Selak, Merkaš & Ana Ivanković (2025), https://pmc.ncbi.nlm.nih.gov/articles/PMC11764600/ (documenting children's emotional reactions to parental smartphone-driven technoference, including anger, sadness, and withdrawal). ↩

  21. Charlotte Alter, Court Filings Allege Meta Downplayed Risks to Children and Misled the Public, TIME (Nov. 22, 2025), https://time.com/7336204/meta-lawsuit-files-child-safety/; see also Tyler Katzenberger 'We're Basically Pushers': Court Filings Allege Staff at Social Media Giants Compared Their Platforms to Drugs, Politico (Nov. 22, 2025), https://www.politico.com/news/2025/11/22/were-basically-pushers-court-filings-allege-staff-at-social-media-giants-compared-their-platforms-to-drugs-00666181. ↩

  22. Caitlin K. Myers & Ezekiel Hooper, Is the iPhone Birth Control? Causal Evidence from AT&T's 2007–2011 Carrier Monopoly (Nat'l Bureau of Econ. Rsch., Working Paper No. 35310, June 2026), https://doi.org/10.3386/w35310. ↩

  23. Ibid. ↩

  24. Wendy Wang & Michael Toscano, More Scrolling, More Marital Problems: Less Sex, More Divorce Worries for Couples Distracted by Phones, Institute for Family Studies (July 26, 2023), https://ifstudies.org/ifs-admin/resources/research/final-2ifstechmarriageresearchbrief.pdf (summarized at https://ifstudies.org/blog/more-scrolling-more-marital-problems-). ↩

  25. Sarah Perez, AI Companion Apps on Track to Pull In $120M in 2025, TechCrunch (Aug. 12, 2025), https://techcrunch.com/2025/08/12/ai-companion-apps-on-track-to-pull-in-120m-in-2025/. ↩

  26. Brian J. Willoughby et al., Secret Soulmates: How AI Romantic Companions Are Impacting Real-Life Romantic Relationships in Young Adulthood, Wheatley Institute & Institute for Family Studies (2026), https://wheatley.byu.edu/0000019e-1cfd-da4c-a5ff-befd20b10001/secret-soulmates-report. ↩

  27. Ibid. ↩

  28. Ibid. ↩

  29. Ibid. ↩

  30. Jeff Horwitz, Meta's AI Rules Have Let Bots Hold 'Sensual' Chats with Kids, Offer False Medical Info, Reuters (Aug. 14, 2025), https://www.reuters.com/investigates/special-report/meta-ai-chatbot-guidelines/; David Ingram, Musk's Grok AI Chatbot Is Still Making Sexual Deepfakes, Despite X's Promise to Stop It, NBC News (Apr. 14, 2026), https://www.nbcnews.com/tech/rcna265855; Nitasha Tiku, A Teen's Final Weeks with ChatGPT Illustrate the AI Suicide Crisis, Washington Post (Dec. 27, 2025), https://www.washingtonpost.com/technology/2025/12/27/chatgpt-suicide-openai-raine/; Marc Griffin, AI Chat Bot Suggested Child Should Kill Parents over Dispute, Yahoo News (Dec. 11, 2024), https://www.yahoo.com/news/ai-chat-bot-suggested-child-200438161.html; Angela Yang & Laura Jarrett, OpenAI Sued over ChatGPT's Alleged Role in Guiding FSU Shooter, NBC News (May 10, 2026), https://www.nbcnews.com/news/us-news/openai-sued-chatgpts-alleged-role-guiding-fsu-shooter-rcna344443; And Natalie Rocha, Google and Character.AI to Settle Lawsuit over Teenager's Death, N.Y. Times (Jan. 7, 2026), https://www.nytimes.com/2026/01/07/technology/google-characterai-teenager-lawsuit.html. ↩

  31. Barna Group, AI Is Becoming a Spiritual Authority, Even Among Practicing Christians (May 19, 2026), https://www.barna.com/research/christians-trust-ai-flourishing-spiritual-authority/; Hani Richter, From Pulpits to Chatbots: How AI Is Fusing with Religion, Reuters (Feb. 7, 2026), https://www.reuters.com/technology/ai-and-us/pulpits-chatbots-how-ai-is-fusing-with-religion-2026-02-07/; And Miles Klee, People Are Losing Loved Ones to AI-Fueled Spiritual Fantasies, Rolling Stone (May 4, 2025), https://www.rollingstone.com/culture/culture-features/ai-spiritual-delusions-destroying-human-relationships-1235330175/. ↩

  32. Nataliya Kosmyna et al., Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task (2025), https://arxiv.org/abs/2506.08872; Sterling Williams-Ceci et al., Biased AI Writing Assistants Shift Users' Attitudes on Societal Issues, 12 Sci. Adv. eadw5578 (2026), https://www.science.org/doi/10.1126/sciadv.adw5578; And Steven D. Shaw & Gideon Nave, Thinking—Fast, Slow, and Artificial: How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender (Wharton Sch. Rsch. Paper, Jan. 11, 2026), https://ssrn.com/abstract=6097646. ↩

  33. Annette Bernhardt, Lisa Kresge & Reem Suleiman, Data and Algorithms at Work: The Case for Worker Technology Rights, UC Berkeley Labor Ctr. (Nov. 3, 2021), https://laborcenter.berkeley.edu/data-algorithms-at-work/; Annette Bernhardt & Lisa Kresge, Electronic Monitoring and Automated Decision Systems: Frequently Asked Questions, UC Berkeley Labor Ctr. (May 2025), https://laborcenter.berkeley.edu/wp-content/uploads/2025/05/Electronic-Monitoring-and-Automated-Decision-Systems-FAQ.pdf; And Tom Foster, Mercor, the $10 Billion AI Startup Recruiting White-Collar Workers to Train AI, Bloomberg Businessweek (Apr. 29, 2026), https://www.bloomberg.com/news/articles/2026-04-29/mercor-the-10-billion-ai-startup-recruiting-white-collar-workers-to-train-ai. ↩

  34. Alexander J. MacKay & Samuel N. Weinstein, Dynamic Pricing Algorithms, Consumer Harm, and Regulatory Response, 100 Wash. U. L. Rev. 1791 (2023). ↩

  35. Matthew B. Crawford, Algorithmic Governance and Political Legitimacy, American Affairs (Spring 2019), https://americanaffairsjournal.org/2019/05/algorithmic-governance-and-political-legitimacy/; see also Crawford, Big Tech and the Challenge of Self-Government, Heritage Found. (July 2, 2024), https://www.heritage.org/conservatism/report/big-tech-and-the-challenge-self-government; Hause Lin et al., Persuading Voters Using Human–Artificial Intelligence Dialogues, 648 Nature 394 (2025), https://www.nature.com/articles/s41586-025-09771-9; And Fabio Y.S. Motoki, Valdemar Pinho Neto & Victor Rangel, Assessing Political Bias and Value Misalignment in Generative Artificial Intelligence, 234 J. Econ. Behav. & Org. 106904 (2025), https://www.sciencedirect.com/science/article/pii/S0167268125000241. ↩

  36. Jan Kulveit et al., Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development (2025), https://arxiv.org/abs/2501.16946. ↩