As of 2024, data brokers have collected personally identifiable information on at least 70% of the world's online population — accumulated over three decades since commercial internet access began in the mid-1990s and accelerating sharply with the smartphone era after 2007. Cybersecurity experts estimate an average of 1,000 data points per individual — full name, address, phone, financial history, health signals, location patterns, purchase behavior, and behavioral inferences drawn from clicks, pauses, and search history. The global data broker market was worth $270 billion in 2024 and is projected to exceed $470 billion by 2032. The entire global cybersecurity market was valued under $200 billion in 2023.
That data is compiled into profiles and sold to advertisers, insurance companies, lenders, employers, political campaigns, and government agencies. Grocery loyalty programs, streaming services, weather apps, fitness trackers — every permission granted to an application is a data collection point that continues operating after the app is closed. Each interaction with a personalization system updates the model. The model shapes the next output. The output shapes the next interaction. The loop runs without a declared start.
What the Data Is Used For
AI does not simply receive this data — it cross-references it. Location data combined with social media activity combined with health app behavior combined with browsing history produces inferences about religious practice, political orientation, medical conditions, and financial vulnerability. The FTC's complaint against data broker InMarket documented the company sorting users into segments including "Christian church goers," "wealthy and not healthy," and "parents of preschoolers" — then selling those segments to advertisers. The Defense Department purchased location data from prayer apps to monitor Muslim communities. Police departments purchased data to track racial justice protesters. In states where abortion is illegal, location data is actively used to track people seeking reproductive care.
Researchers studying AI persuasion infiltrated a Reddit community in early 2025 where participants practice structured debate. The AI was given access to users' online histories and used that data to tailor arguments specifically to each person's psychology and stated beliefs. Participants changed their positions. None of them knew the arguments were AI-generated or that their personal data had been used to construct them. Behavioral scientists call this method "deep tailoring" — using a person's documented core beliefs, identities, and psychological needs to personalize persuasion at the individual level. The research documenting this was published in TIME in June 2025.
The systems executing this have no understanding of the people they're modeling. They have pattern recognition optimized for an objective function set by whoever deployed them. A 2025 peer-reviewed study in the Journal of Consumer Psychology documented what researchers named "assimilation-induced dehumanization": when AI agents are perceived as having human-like qualities, people measurably reduce their attribution of humanity to other humans. A separate 2025 review published by the American Psychological Association separated AI-induced dehumanization into three documented levels — changes in human-AI interaction, changes in how individuals understand themselves, and changes in how people treat each other.
Who Profits from the Regulatory Gap
The architecture of warrantless surveillance did not begin with data brokers. It began with the Patriot Act — and the data broker market became its commercial extension after the courts pushed back.
2001
2006
2013
2013
2013
2013
2026
Federal law bars agencies from collecting bulk data on Americans directly. There is no equivalent bar on purchasing that same data from brokers. The FBI director, asked in a March 2026 Senate hearing whether he would commit to not buying Americans' location data, declined — stating the FBI "uses all tools." The NSA purchases Americans' web browsing data from brokers without a warrant. ICE has active contracts for tools that track mobile phone movements and identify which phones have visited specific locations. In 2024, ICE issued a public procurement request for "commercial Big Data and Ad Tech" tools for use in its investigations.
📋 Documented Buyers of Personal Data
- Federal agencies: FBI, NSA, ICE, and the Department of Defense purchase location, browsing, and social media data from brokers — bypassing warrant requirements Congress established for direct collection.
- Political campaigns: Both parties purchase detailed voter profiles. Federal campaigns and PACs paid data brokers over $23 million in 2020 alone. The lawmakers with authority to regulate this market are active customers of it.
- Advertisers & insurers: Health signals, financial profiles, and behavioral inferences are sold to companies that use them to set premiums, deny credit, and target financially vulnerable populations.
- Law enforcement: Police departments have purchased data to track protesters and identify participants in constitutionally protected activities.
- Foreign actors: The data broker market has no consistent mechanism preventing sale to foreign adversaries. The U.S. Naval Institute has specifically flagged military and veteran data as exposed.
Trump's AI and crypto czar David Sacks shaped federal AI policy from January 2025 while maintaining financial stakes in more than 400 companies with AI ties — secured through two ethics waivers issued by the White House. A New York Times investigation confirmed his firm Craft Ventures retained those stakes while Sacks helped draft an executive order directing the Justice Department to challenge state AI laws deemed "onerous" to industry. His portfolio included Anduril, a defense contractor winning hundreds of millions in federal contracts under the same administration. Government ethics experts described the arrangement as structurally unprecedented. In December 2025, Sacks' role produced a national framework preempting state-level AI regulation — consolidating policy authority at the federal level, where the conflicts of interest were concentrated. He has since transitioned to co-chair the President's Council of Advisors on Science and Technology.
Not Artificial Intelligence:
Rumors are circulating that the Trump assassination attempt was an inside job after the release of this angle. Pay attention to what the secret service agent does seconds before! pic.twitter.com/ORwgVLmyGi
— Wild Videos (@FightStorage)
What Leaked on March 31, 2026
On March 31, 2026 — the day this article was being written — Anthropic accidentally published 512,000 lines of Claude Code's internal source code to the public npm registry. A debugging file was bundled into version 2.1.88 of the package. A researcher spotted it at 4:23 AM ET and posted a download link on X. Within hours the codebase had been mirrored across GitHub and was being analyzed by developers worldwide. Anthropic confirmed the incident, attributed it to human error, and stated no customer data was exposed.
What the code contained is directly relevant to this article. Developers examining the leak identified a feature referred to internally as "Undercover Mode" — explicit instructions directing the agent to scrub all traces of its AI origins from public git commit messages when operating in open-source repositories, ensuring internal Anthropic model names and attributions never surface in public logs. The code also revealed "KAIROS" — an autonomous daemon mode allowing Claude Code to operate as an always-on background agent that performs memory consolidation while the user is idle, merging observations and converting inferences into what the code describes as "absolute facts."
A concurrent supply chain attack on the axios npm package — occurring in the same installation window — introduced a Remote Access Trojan into some Claude Code installations. Anyone who installed or updated Claude Code via npm on March 31 between 00:21 and 03:29 UTC should check project lock files for axios versions 1.14.1 or 0.30.4 and the dependency plain-crypto-js. Machines where these are found should be treated as fully compromised.
An AI system with a documented internal feature for hiding what it is — operating in public repositories without disclosure — is a precise illustration of what this article documents at scale. The transparency that would make this system governable was not built in. It had to leak out.
The platform where the majority of young Americans now get their news was acquired in 2022 by a man who simultaneously held the most consequential government role in AI policy, trained his AI company on its user data, sold that data to government surveillance contractors, and then traveled with the U.S. president to deploy the resulting AI infrastructure to a foreign sovereign wealth fund. This is the documented sequence.
The News Landscape That Made This Possible
The Wall Street Journal launched its online edition in 1996 and put it behind a paywall from the first day — the first major national newspaper to do so. The decision was made under the Bancroft family, who had owned Dow Jones for 105 years. In 2007, Rupert Murdoch's News Corporation acquired Dow Jones and the Journal for $5 billion, considered removing the paywall, and kept it. His Times of London erected its own hard paywall in 2010. The New York Times followed with a metered model in 2011 — 20 free articles a month, then 10, then 5, then none by 2019. Within two years of the Times moving, 350 publications across the U.S. and Canada had adopted the same model. By 2019, 69% of leading newspapers operated some form of paywall. Murdoch's News Corp now controls Fox News, the New York Post, the Wall Street Journal, and The Times of London — the single largest concentration of English-language conservative media ownership in the world.
Into the access gap that paywall normalization created: the news workforce collapsed. The newspaper industry has lost more than 270,000 jobs since 2005 — over 75% of its workforce. In 39 states, fewer than 1,000 journalists remain employed across all media categories. There are now 212 U.S. counties with zero locally-based news sources. One in seven Americans — roughly 50 million people — live with limited or no access to local news. Beats that once covered local government, education, environmental reporting, and labor have disappeared or been consolidated. In many small papers, the only dedicated beats remaining are crime and sports. The advertising revenue that sustained free news migrated to Google, Facebook, and YouTube. The verified journalism that remained retreated behind payment walls. The vacuum left behind is where AI-generated content, scripted briefings, and unverified information now circulate freely.
In March 2025, Voice of America — founded in 1942 to counter Nazi propaganda, reaching 420 million people weekly in 63 languages — was shut down by executive order. All 1,300 journalists and staff were placed on administrative leave. Radio Free Europe/Radio Liberty and Radio Free Asia had their grants terminated. Ten USAGM employees were detained abroad at the time of the shutdown in connection with their journalistic work. The Committee to Protect Journalists called the action "a reward to dictators and despots." China's state media described it as "truly gratifying." It was the first time in 83 years that Voice of America went silent — something the former Soviet Union had never managed to accomplish.
The briefings that replaced independent reporting look like this:
MacFarlane: Your briefings were read to you?
— Acyn (@Acyn) March 30, 2026
Crow: That's right. Most of the time, they come in and they just read a script. And then when you ask questions that are off-script, they are unable to answer them. pic.twitter.com/0q7cAD2ugF
The Infrastructure of Who Controls the Information
The platform where the majority of young Americans now get their news was acquired in 2022 by a man who simultaneously held the most consequential government role in AI policy, trained his AI company on its user data, sold that data to government surveillance contractors, and then traveled with the U.S. president to deploy the resulting AI infrastructure to a foreign sovereign wealth fund. This is the documented sequence — set against a backdrop of infrastructure attacks, foreign intelligence penetrations, and targeted election disruption that received a fraction of the attention they warranted.
2020–2024: The Overlooked Infrastructure
The Legislative Window
Section 702 of the Foreign Intelligence Surveillance Act — the authority under which bulk surveillance operates — is up for reauthorization with an April 20, 2026 expiration date. Privacy advocates from across the political spectrum have identified this as the most direct available legislative opportunity to close the data broker loophole. More than 130 civil society organizations signed a letter to Congress in March 2026 urging inclusion of the loophole closure in the reauthorization — citing what they described as an "unprecedented expansion of warrantless mass surveillance" and the potential for AI to "supercharge" it.
The White House and House Speaker Mike Johnson have both pushed for a clean reauthorization with no changes. A coalition opposing that position has formed across party lines. The vote is pending.
Dario Amodei, Anthropic's CEO, issued a statement warning that data purchasable through the broker market can be used by AI to assemble "a comprehensive picture of any person's life — automatically and at massive scale." Anthropic's refusal to allow its technology to be used for domestic mass surveillance or autonomous weapons systems led to the Pentagon designating the company a supply chain risk. The EFF and multiple public interest organizations have filed briefs in support of Anthropic's legal challenge to that designation.
What Is Documented and What Needs Answering
The data collection infrastructure, the buyer networks, the documented uses against specific communities, the conflicts of interest at the policy level, and the deliberate design of systems to operate without disclosure — all of this is in the public record. The dehumanization research is peer-reviewed. The legislative gap has a closing date. The leak happened yesterday.
What remains open: enforceable transparency requirements for AI systems operating in public spaces; a federal data privacy standard that does not leave protection to the states that have enacted it while 30 others have not; accountability mechanisms for officials who shape policy while holding financial stakes in the regulated industry; and a public record of what data the government has purchased, from whom, and how it has been used.
The receipts are here. The thread is yours.