AI

The Concentration Engine: How AI in the Hands of the Few Threatens the Many

AI is a wonderful tool, but if left unchecked in the wrong hands, “the people” won’t stand a chance.

In 2024, corporate investment in artificial intelligence reached $252.3 billion. In 2025 it more than doubled to roughly $581 billion globally, with $344 billion of it flowing into the United States [1]. Over 90% of "notable" AI models released in 2025 came from private industry, almost none from academia or government [1]. A handful of firms (OpenAI, Anthropic, Google DeepMind, Meta, xAI) dominate frontier model development, and the compute beneath them is more concentrated still: AWS, Microsoft Azure, and Google Cloud together accounted for 63% of enterprise cloud infrastructure spending in the third quarter of 2025 [2].

Global freedom has now declined for twenty consecutive years; internet freedom for fifteen [7]. People are growing concerned about what AI will do to accelerate this asymmetry. Even Pope Leo XIV, the first American pope, has made the concentration of AI power a defining moral question of his pontificate, asking in his December 5, 2025 address to the Strategic Alliance of Catholic Research Universities: "How can we ensure that the development of artificial intelligence truly serves the common good, and is not just used to accumulate wealth and power in the hands of a few?" [8].

This article argues that the existential threat of AI is not, in the first instance, the science-fiction scenario of rogue machines. It is the more banal threat of concentrated power: governments and large corporations using AI to entrench dominance over ordinary citizens in labor markets, in political discourse, in surveillance, and in the means of lethal force. It targets the slow, almost imperceptible concentration of power which may, in the future, rapidly accelerate. We believe the remedy is not to ban AI, nor to surrender to a regulatory regime crafted by incumbents. AI can be used for pro-democracy and pro-social ends as well. We can fight back with AI. So instead, we argue that the remedy is Title II of the MAD Act, referred to as "Demand A Plan for AI," a congressionally mandated investigation modeled on the proven phased frameworks of past landmark legislation [9], paired with targeted immediate protections and automatic enforcement triggers.

I. The Economics of Concentration: Who Owns the Frontier

"AI tends to amplify the power of those who already possess economic resources, expertise and access to data." Pope Leo XIV, Magnifica Humanitas, May 25, 2026 [10].

Why is this important to do? Let’s first look at the numbers. According to Stanford's 2025 AI Index Report, U.S. private AI investment hit $109.1 billion in 2024, nearly twelve times China's $9.3 billion and twenty-four times the United Kingdom's $4.5 billion [1]. The 2026 update finds that 2025 set a new record at roughly $581 billion in global AI investment, with $344 billion in the United States [1]. Generative AI alone attracted $33.9 billion in private investment in 2024, more than eight times the 2022 level [1].

But frontier model development is concentrated in a handful of firms. In 2025, over 90 percent of "notable" AI models came from private companies, with only a small share from academia [1]. OpenAI, Anthropic, Google DeepMind, Meta, and xAI dominate, while Microsoft, Amazon, and Google have written checks of historic size to maintain stakes in those companies they do not own outright. In 2025, Google committed up to $40 billion to Anthropic, with an initial $10 billion at a $350 billion valuation and the remainder contingent on milestones, an investment that followed Amazon's parallel $5 billion to $20 billion commitment [11]. By late 2026, Anthropic had raised a $30 billion Series G at a $380 billion post-money valuation, with annualized revenue rising from $1 billion at the end of 2024 to $30 billion in early 2026, a roughly thirty-fold increase in fifteen months [11].

The compute layer is more concentrated still. Synergy Research Group reports that Amazon, Microsoft, and Google together accounted for 63 percent of enterprise spending on cloud infrastructure services in the third quarter of 2025, a market of roughly $107 billion for that quarter alone [2]. Training compute for frontier models has been doubling roughly every five months [1]. A single firm, TSMC, fabricates the majority of advanced AI chips. The AI stack is therefore a vertically integrated oligopoly: the same firms that own the foundational chips own the cloud, own equity in the model labs, and increasingly own the application layer that touches consumers.

This is what economists used to call a trust, and what Justice Louis Brandeis, who spent his career fighting trusts, warned would prove incompatible with democratic government itself. (The famous Brandeis line, "We can have democracy in this country, or we can have great wealth concentrated in the hands of a few, but we can't have both," is, we should note, of disputed primary-source provenance. Legal scholar Peter S. Campbell of the Green Bag exhaustively searched Brandeis's writings and found no exact match, though scholars concede the sentiment is consistent with Brandeis's thinking [12]. We flag the attribution rather than presenting it uncritically.)

Again we point out how Pope Leo XIV captured the structural point at the very moment of his election. In his first formal address to the College of Cardinals on May 10, 2025, two days after assuming the chair of Peter, the new Leo XIV told the assembled cardinals: "I chose to take the name Leo XIV. There are different reasons for this, but mainly because Pope Leo XIII in his historic Encyclical Rerum Novarum addressed the social question in the context of the first great industrial revolution. In our own day, the Church offers to everyone the treasury of her social teaching in response to another industrial revolution and to developments in the field of artificial intelligence that pose new challenges for the defence of human dignity, justice and labour" [13].

A year later, in his first encyclical Magnifica Humanitas, released May 25, 2026, Leo set the issue at the center of his pontificate, framing it as a "culture of power" question and warning that AI must not be allowed to accumulate wealth and authority in the hands of a few [10].

That is a central question we ourselves must pontificate on, and plan for.

II. Labor, the Permanent Underclass, and the Hollowing of the Middle

"This 'useless class' will not be merely unemployed, it will be unemployable." Yuval Noah Harari, Homo Deus: A Brief History of Tomorrow (2017) [14].

A part of the asymmetry that AI may exacerbate will be concerning economic power. When IMF Managing Director Kristalina Georgieva addressed the World Economic Forum in Davos in January 2024 and again in January 2026, she did not soften the language. "This is like a tsunami hitting the labor market," she said [15]. IMF research projects that AI will affect 60%of jobs in advanced economies and approximately 40% globally; roughly half of exposed workers may see productivity gains, while the other half face displacement, wage suppression, or outright disappearance of their roles [3]. "Wake up," Georgieva said at Davos 2026. "AI is for real, and it is transforming our world faster than we are getting a handle on" [15].

The pattern of displacement is not random. It is structural.

The McKinsey Global Institute projects that, by 2030, activities that account for up to 30% of hours currently worked across the U.S. economy could be automated, a trend McKinsey says is accelerated by generative AI [4]. The report notes: "Some 8.6 million occupational shifts took place from 2019 through 2022. Now even more change is in store. We expect an additional 12 million occupational shifts by 2030" [4]. McKinsey finds the affected workers fall disproportionately into office support, customer service, and food service occupations [4].

Goldman Sachs economists Joseph Briggs and Devesh Kodnani estimated in March 2023 that "generative AI could expose the equivalent of 300mn full-time jobs to automation," with roughly two-thirds of current U.S. jobs exposed to some degree of AI automation and generative AI capable of substituting up to one-fourth of current work [5].

The Anthropic Economic Index, a large-scale dataset built from how AI is actually being used and mapped against the U.S. Department of Labor's O*NET task database, finds that AI use is concentrated in software development and technical writing, with computer and mathematical tasks the single largest share of Claude queries [16]. Anthropic's recent index reports show automation rising on its API platform, particularly in business sales and outreach workflows and customer service [16].

Georgieva, speaking at Davos 2026, added that "the middle class, inevitably, is going to be affected" [15]. This is what labor economists have begun calling hiring suppression: not mass layoffs of incumbents, but the silent elimination of the entry-level rung of the ladder. Goldman Sachs Research's 2025 follow-up reported that unemployment among workers aged 20 to 30 in tech-exposed occupations had risen meaningfully relative to peers in less-exposed trades, an early empirical signature of the shift [17].

Yuval Noah Harari has named the projected outcome the "useless class," a phrase from his 2016 book Homo Deus and elaborated in 21 Lessons for the 21st Century [14]. Harari's point is not that human beings are useless; it is that the market may decide they are. "This 'useless class' will not be merely unemployed, it will be unemployable," he wrote [14]. The risk is the modern version of what Marx called a reserve army of labor: a permanently surplus population kept docile by what Harari has elsewhere described as a combination of basic income, drugs, and computer games.

The Vatican's January 28, 2025 doctrinal note Antiqua et Nova, co-issued by the Dicasteries for the Doctrine of the Faith and for Culture and Education, addresses this directly: "while AI promises to boost productivity, current approaches to the technology can paradoxically deskill workers, subject them to automated surveillance, and relegate them to rigid and repetitive tasks" [18]. The document, 117 paragraphs signed by Cardinals Víctor Manuel Fernández and José Tolentino de Mendonça, expresses concern that AI's benefits have flowed primarily to a small number of dominant companies.

It is worth pausing to note what is unprecedented. In previous industrial revolutions, many displaced workers eventually found new work elsewhere in the economy. Human hands were still needed elsewhere. MIT economist David Autor and co-authors, in a study of eighty years of U.S. census data, found that the majority of U.S. workers today are in occupations that have only emerged widely since 1940 [19]. But the previous transitions distributed gains widely; the AI transition, so far, distributes them narrowly.

A class that is struggling to find ways to feed, clothe, and shelter itself will scarcely have the time and resources to fight back at the erasure of their rights.

III. The Surveillance State and Democratic Backsliding

"The ideal subject of totalitarian rule is not the convinced Nazi or the convinced Communist, but people for whom the distinction between fact and fiction (i.e., the reality of experience) and the distinction between true and false (i.e., the standards of thought) no longer exist." Hannah Arendt, The Origins of Totalitarianism (1951), Part 3, Chapter 13 [20].

Freedom House's Freedom in the World 2026 report finds that global freedom has now declined for the twentieth consecutive year, with deterioration in dozens of countries and improvement in far fewer [7]. Just a fraction of the world's people now live in countries rated Free, down sharply from two decades ago [7]. The United States itself, the country that wrote the playbook on AI, experienced one of the sharpest annual declines of any country rated Free in 2025 [7].

The mechanism is, in important part, technological. Freedom House's Freedom on the Net 2025 parallel index finds that global internet freedom has declined for fifteen consecutive years, with conditions deteriorating in dozens of countries [7]. AI-based content moderation, AI surveillance, and AI-driven content fabrication have together produced what Freedom House describes as a significant escalation of disinformation and rights-undermining practices globally [7].

The asymmetry of AI surveillance is staggering. Per Clearview's own company statements, reported by Nextgov/FCW in March 2025, Clearview AI's database now contains roughly 60 billion facial images scraped from social media [21]. NSO Group's Pegasus spyware, according to the seventeen-outlet Pegasus Project journalism consortium of July 2021, was used to attempt to hack smartphones belonging to journalists, human-rights activists, business executives, and government officials across approximately fifty countries [22]. China's social credit system, predictive policing, and pervasive facial recognition are exported to authoritarian regimes worldwide. Palantir's Maven Smart System, its AI battlefield-targeting and intelligence-aggregation platform, saw its U.S. Army contract ceiling raised by $795 million in May 2025, bringing the total contract value to roughly $1.3 billion through 2029 [23]. In July 2025, the Army awarded Palantir a separate $10 billion enterprise agreement consolidating dozens of software and data contracts [23].

Inside the United States, federal authorities have announced plans to deploy AI-assisted social-media surveillance systems to monitor the online activity of student visa recipients, generating high-profile detentions of foreign nationals after visa revocations based on online expression [7].

This is the architecture Hannah Arendt described in another century: the slow erasure of the distinction between fact and motive, between public and private, between citizen and suspect. In a 1967 essay later collected in Between Past and Future, she wrote: "The result of a consistent and total substitution of lies for factual truth is not that the lie will now be accepted as truth and truth be defamed as a lie, but that the sense by which we take our bearings in the real world... is being destroyed" [24].

The most dangerous fact about AI-enabled surveillance is not what it allows the state to do; it is what it allows the state to know the people are doing. The asymmetry between governor and governed becomes total when every email, location ping, and search query feeds an aggregated model. How can resistance coordinate when it is being watched by a government that has a long history of quieting dissent?

IV. Manufactured Consent at Machine Scale

"Mass propaganda discovered that its audience was ready at all times to believe the worst, no matter how absurd, and did not particularly object to being deceived because it held every statement to be a lie anyhow." Hannah Arendt, The Origins of Totalitarianism (1951) [20].

There is more than one way to skin a cat. Quieting dissent can be via physical interventions in coordination between groups, or it can arise through the subtle manipulation of the information environment which pacifies the public. In 1988, Edward S. Herman and Noam Chomsky published Manufacturing Consent (we know, we know, he’s in the Epstein files), arguing that the structural ownership and advertising-funding of mass media filtered news in favor of elite interests without any need for overt censorship [25]. The propaganda model worked because the filters (ownership concentration, advertiser pressure, sourcing dependence, ideological flak) operated invisibly inside ostensibly free institutions.

AI does not merely accelerate that model. It personalizes it.

In March 2026, the National Republican Senatorial Committee released an AI-generated deepfake video of Texas Democratic Senate candidate James Talarico, a real-looking, computer-fabricated version of the candidate that appeared to speak directly into the camera for more than a minute [26]. This was not a marginal experiment; it was the official Senate campaign arm of a major American political party, in a U.S. Senate race.

And it is not the only experiment. In Slovakia's 2023 parliamentary elections, deepfake audio purporting to capture liberal candidate Michal Šimečka discussing electoral fraud surfaced 48 hours before voting, during the legally mandated silence period. Šimečka lost; the pro-Russian populist Robert Fico won. Harvard Kennedy School's Misinformation Review cautions against attributing Fico's victory to the deepfake alone, but the incident demonstrated the genre's potential [27]. In the United States, the congressional findings underlying Demand A Plan for AI cite survey research showing strong increases in voter skepticism about online content as a consequence of deepfakes, the so-called "liar's dividend" that Bobby Chesney and Danielle Citron warned about as early as 2018: when anything can be fake, anything real can be dismissed as fake [28].

Foreign actors have not been idle. Freedom House documents dozens of governments that have deployed online commentators and AI-augmented influence operations to manipulate discussions in their favor, a sharp increase from a decade ago [7]. Russia, China, and Iran have all run AI-augmented influence operations in U.S. and European elections. The Cambridge Analytica scandal of 2018 prefigured the model; today's microtargeting is exponentially more precise.

The economic incentive structure makes this worse. The cost of creating a fake is marginal, especially for the elite; the cost of debunking it is high; and the time asymmetry between viral spread and forensic verification is fatal. Speaking of fatal -

V. Autonomous Weapons and the Mechanization of Killing

"Artificial Intelligence now demands to be disarmed, freed from logics that turn it into an instrument of domination, exclusion and death." Pope Leo XIV, presenting Magnifica Humanitas, May 25, 2026 [29].

The clearest case of AI's transfer of decisional weight from human to system, and from citizen to state, is in lethal autonomy.

Between October 2023 and 2024, the Israel Defense Forces' AI targeting systems, "Lavender," "The Gospel" (Habsora), and "Where's Daddy?", were revealed in a six-officer-sourced investigation by journalist Yuval Abraham of +972 Magazine and Local Call, with reporting confirmed by The Guardian [30]. According to the +972 reporting, Lavender at one point marked as many as 37,000 Palestinians as suspected militants for possible air strikes [30]. Sources told +972 that Lavender's error rate was acknowledged at roughly 10 percent; human review of its targets was, in one source's words, "a rubber stamp" lasting on average about "20 seconds," "just to make sure the Lavender-marked target is male" [30]. "Where's Daddy?" tracked flagged individuals until they returned to their family homes, at which point strikes were authorized [30]. As one intelligence source put it to +972: "You put hundreds [of targets] into the system and wait to see who you can kill" [30].

These targeting systems ran in part on commercial cloud infrastructure. The Guardian reported that Israeli signals intelligence Unit 8200 had been storing very large volumes of intercepted Palestinian phone traffic on Microsoft Azure, with leaked Microsoft documents referencing thousands of terabytes of Israeli military data on Azure servers in Europe [31].

Inside the United States, Project Maven, originally controversial enough that thousands of Google employees signed a 2018 protest letter that led Google to decline renewal, has since matured into Palantir's billion-dollar Maven Smart System described above [23]. Anduril Industries has scaled rapidly with contracts for autonomous "Collaborative Combat Aircraft." OpenAI and Anthropic, despite their stated safety missions, have established defense partnerships of various scopes. The category of "the AI safety company" and the category of "the defense contractor" are merging.

The Vatican's Antiqua et Nova, paragraphs 100 and 101, warns that autonomous lethal weapons systems "capable of identifying and striking targets without direct human intervention are a cause for grave ethical concern," and that such systems, in Pope Francis's words quoted in the document, pose "an 'existential risk' by having the potential to act in ways that could threaten the survival of entire regions or even of humanity itself" [18]. The International Committee of the Red Cross has called for prohibitions on fully autonomous weapons. The Stop Killer Robots coalition, drawing on United Nations CCW Group of Governmental Experts processes since 2014, has pressed for binding treaty obligations. Stuart Russell, professor of computer science at UC Berkeley and co-author of the field's standard textbook, has warned that lethal autonomous weapons constitute "the third revolution in warfare, after gunpowder and nuclear arms" [32].

In May 2023, the Center for AI Safety published a one-sentence statement signed by hundreds of executives, researchers, and engineers, including Geoffrey Hinton, Yoshua Bengio, Sam Altman, Demis Hassabis, Dario Amodei, and Bill Gates: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war" [33]. Hinton, who had resigned from Google days earlier specifically to speak freely about the risks, told The New York Times: "I console myself with the normal excuse: If I hadn't done it, somebody else would have" [34]. He added: "It is hard to see how you can prevent the bad actors from using it for bad things" [34].

The UK AI Security Institute's December 2025 Frontier AI Trends Report finds that frontier model capabilities are doubling roughly every eight months, with AI agent task-completion rates on long-horizon software tasks rising sharply from a near-zero baseline only two years earlier, and self-replication success rates rising from low single digits in 2023 to a majority by 2025 in simplified evaluations [6].

This is the speed at which power is being delegated to machines that ordinary citizens neither own, audit, nor understand.

VI. The Counterpoints, Honestly Considered

A serious case must engage its strongest counterarguments. There are three worth naming.

First, AI offers real and distributed benefits. GPT-3.5-equivalent performance dropped in cost by orders of magnitude between November 2022 and late 2024 [1]. FDA-approved AI-enabled medical devices grew from single digits in 2015 to several hundred by 2023 [1]. Coding assistants have measurably narrowed productivity gaps between junior and senior developers. AI tutors are reaching students in countries where teachers are scarce. These gains are not hypothetical, and any policy regime that crushes them is morally suspect.

Second, heavy-handed regulation could entrench incumbents. This is the genuine fear of the open-source community, of Yann LeCun at Meta, and of many small developers. A regulatory regime designed by the largest labs will, predictably, favor the largest labs. Compliance costs that are trivial for a $380 billion company are existential for a five-person startup. The EU AI Act's risk-tiered framework, well-intentioned, has been criticized on exactly these grounds. California's SB-1047, vetoed in 2024, and its successor bills face the same critique. There is a real risk that "safety" becomes the moat behind which the present oligopoly fortifies itself.

Third, open-weight models are a democratic counterweight. Meta's Llama family, Mistral's open releases, China's DeepSeek and Alibaba's Qwen models all reduce the moat that pure proprietary scale would otherwise produce. The UK AISI notes that open-weight models are particularly difficult to defend against misuse, but that same difficulty is also their democratic virtue: they cannot be silently revoked, audited away, or weaponized by a single firm [6].

These considerations are not refutations of the concentration thesis. They are reasons for calibrated policy rather than panicked policy. The danger runs in two directions at once: overregulation that strangles individual liberty, entrepreneurial dynamism, and the genuine democratizing potential of distributed AI tools, and deregulatory drift that lets harms spiral, concentrates power further, harms the economy, and accelerates the unchecked militarization of AI. A wrong-shape regulation may be worse than no regulation. A right-shape regulation requires, first, an honest investigation of who actually holds the levers and where the harms are actually landing.

That is precisely what Title II of the MAD Act proposes [9].

VII. The Policy Call: Pass Title II of the MAD Act

Title II of the MAD Act, "Demand A Plan for AI," mandates a comprehensive congressional investigation into AI's economic, democratic, military, and societal effects before sweeping prescriptive regulation [9]. Title II:

1. Models its approach on the proven phased, investigatory legislative strategies Congress used in the Water Quality Act of 1965 (Public Law 89-234) [35] and the Air Quality Act of 1967 (Public Law 90-14) [36], statutes that addressed complex, fast-moving technical domains by funding rigorous, time-bound fact-finding before binding standards were locked in. Those frameworks gave the federal government the empirical record it needed to write rules that actually worked, while resisting the twin temptations of premature regulation and indefinite inaction. The principle is evidence-based policymaking: regulate where harm is documented and consensus exists; investigate where evidence is contested or incomplete; and refuse both extremes, the overregulation that strangles individual liberty and the responsible deployment of beneficial technology, and the deregulatory drift that allows concentration to spiral, the economy to be hollowed out, and dangerous warfare to be accelerated [9].

2. Imposes targeted immediate protections in domains where harm is already documented and consensus exists: political deepfake disclosure rules building on the TAKE IT DOWN Act (Public Law 119-12) [37], election-integrity provisions, prohibitions on non-consensual intimate imagery, child-safety protections against AI companion systems, and red-line prohibitions on autonomous weapons without meaningful human oversight, AI-assisted CBRN weapons development, and AI-generated child sexual abuse material [9].

3. Includes "hammer" provisions that automatically trigger if Congress fails to act on the investigation's findings within a defined window, so the bill cannot be slow-walked into irrelevance by industry lobbying [9].

This is the right architecture for three reasons.

First, it avoids regulatory capture by demanding empirical input before rule-writing. The history of the Telecommunications Act of 1996, the Dodd-Frank reforms, and the early privacy frameworks is a history of rules drafted in the interests of incumbents. A statutorily mandated, time-bound, transparent investigation makes that capture harder.

Second, it respects the unsettled state of the science. The UK AISI report observes that AI capabilities are doubling roughly every eight months; the policy question of 2025 may be obsolete by 2027 [6]. Locking in 2025-era rules guarantees fighting the last war while obstructing the next generation of beneficial deployment.

Third, it walks the fine line that all serious technology governance must walk. Overregulation harms individual liberty, smothers small developers, and entrenches the very oligopoly we are trying to constrain. Underregulation lets a handful of firms and agencies accumulate the means of economic, political, and lethal control before any democratic check can form. The MAD Act is calibrated specifically against both failure modes: immediate protection where harm is documented; investigation where evidence is contested; automatic enforcement triggers if Congress refuses to act on what the investigation finds. The deepest question is not whether models are "safe." It is whether power is balanced.

This approach is a refusal to allow a few hundred billionaires, a handful of corporations, and the agencies of any single government, democratic or otherwise, to control the technology that will shape the next century's economy, war, and discourse. Power tends to corrupt people in many ways, and one of those ways it is persuades those who hold it that they must simply know best. That their might makes them right. But this has always been a fallacy. However, this true does not disarm the powerful of their power, so the truth then doesn’t matter, and as powerful elites can reshape the world in their image, then slowly the truth erodes and gives way to their remaking of the world. We cannot let this happen.

Title II of the MAD Act will not, by itself, prevent AI's concentration. But it is the first honest step: investigate, disclose, and structure the deliberation that must follow - which must ultimately end up being codified into law.

References

If you have found any errors in our reasoning or in references, please bring it to our attention immediately.

[1] Stanford Institute for Human-Centered Artificial Intelligence (HAI), AI Index Report 2025 and AI Index Report 2026. Stanford University. https://aiindex.stanford.edu

[2] Synergy Research Group, "Cloud Market Share Trends: Big Three Together Hold 63% while Oracle and the Neoclouds Inch Higher," November 19, 2025. https://www.srgresearch.com/articles/cloud-market-share-trends-big-three-together-hold-63-while-oracle-and-the-neoclouds-inch-higher. Coverage: Lightwave Online (https://www.lightwaveonline.com/home/article/55337257/amazon-microsoft-and-google-command-63-of-the-cloud-infrastructure-services-market); The Register, "AWS under pressure as big three battle to eat the cloud market," November 20, 2025 (https://www.theregister.com/2025/11/20/aws_loses_market_share_azure_google/).

[3] Mauro Cazzaniga, Florence Jaumotte, Longji Li, Giovanni Melina, Augustus J. Panton, Carlo Pizzinelli, Emma J. Rockall, and Marina M. Tavares, "Gen-AI: Artificial Intelligence and the Future of Work," IMF Staff Discussion Note SDN/2024/001, International Monetary Fund, January 2024. https://www.imf.org/en/Publications/Staff-Discussion-Notes/Issues/2024/01/14/Gen-AI-Artificial-Intelligence-and-the-Future-of-Work-542379

[4] McKinsey Global Institute, Generative AI and the Future of Work in America, July 26, 2023. https://www.mckinsey.com/mgi/our-research/generative-ai-and-the-future-of-work-in-america

[5] Joseph Briggs and Devesh Kodnani, "The Potentially Large Effects of Artificial Intelligence on Economic Growth," Goldman Sachs Global Economics Analyst, March 26, 2023. https://www.goldmansachs.com/insights/articles/generative-ai-could-raise-global-gdp-by-7-percent

[6] UK AI Security Institute (AISI), Frontier AI Trends Report, December 2025. https://www.aisi.gov.uk

[7] Freedom House, Freedom in the World 2026 and Freedom on the Net 2025. https://freedomhouse.org

[8] Pope Leo XIV, "Address to Participants in the Conference 'Artificial Intelligence and Care for Our Common Home' Organized by the Centesimus Annus Pro Pontifice Foundation and the Strategic Alliance of Catholic Research Universities," December 5, 2025. https://www.vatican.va/content/leo-xiv/en/speeches/2025/december/documents/20251205-conferenza.html

[9] The MAD Act, Title II, "Demand A Plan for AI." Statutory text on file with the office of the sponsor.

[10] Pope Leo XIV, Encyclical Letter Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, signed May 15, 2026; released May 25, 2026. https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html

[11] CNBC and NBC News coverage of the Google to Anthropic and Amazon to Anthropic investments, 2024 to 2026; Anthropic Series G announcement materials.

[12] Peter S. Campbell, "Democracy v. Concentrated Wealth: In Search of a Louis D. Brandeis Quote," The Green Bag 16, no. 3 (Spring 2013): 251 to 256.

[13] Pope Leo XIV, "Address of the Holy Father to the College of Cardinals," May 10, 2025. https://www.vatican.va/content/leo-xiv/en/speeches/2025/may/documents/20250510-collegio-cardinalizio.html

[14] Yuval Noah Harari, Homo Deus: A Brief History of Tomorrow (New York: HarperCollins, 2017); excerpted at ideas.ted.com under the title "The Rise of the Useless Class." See also Harari, 21 Lessons for the 21st Century (Spiegel & Grau, 2018).

[15] Kristalina Georgieva, remarks at the World Economic Forum, Davos, January 2024 and January 2026. Coverage in Fortune (January 2024 and January 2026), Yahoo Finance, and Reuters. IMF Managing Director's remarks accompanying release of IMF Staff Discussion Note SDN/2024/001 (see [3]).

[16] Anthropic, The Anthropic Economic Index. https://www.anthropic.com/economic-index

[17] Goldman Sachs Research, follow-up analysis on AI labor-market effects, 2025.

[18] Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education, Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence, January 28, 2025. Vatican summary: https://www.vaticannews.va/en/vatican-city/news/2025-01/new-vatican-document-examines-potential-and-risks-of-ai.html

[19] David Autor, Caroline Chin, Anna Salomons, and Bryan Seegmiller, "New Frontiers: The Origins and Content of New Work, 1940 to 2018," Quarterly Journal of Economics 139, no. 3 (August 2024); NBER Working Paper No. 30389.

[20] Hannah Arendt, The Origins of Totalitarianism (New York: Harcourt, Brace and Company, 1951). Part 3, Chapter 13.

[21] Clearview AI company statements reported in Nextgov/FCW, March 14, 2025.

[22] "The Pegasus Project," Forbidden Stories and seventeen partner outlets including The Washington Post, The Guardian, Le Monde, July 2021.

[23] DefenseScoop coverage of Palantir Maven Smart System contract increase, May 23, 2025; Army announcement of $10 billion Palantir enterprise software agreement, July 2025. https://defensescoop.com

[24] Hannah Arendt, "Truth and Politics," The New Yorker, February 25, 1967; reprinted in Between Past and Future: Eight Exercises in Political Thought (New York: Viking, 1968; revised Penguin, 2006).

[25] Edward S. Herman and Noam Chomsky, Manufacturing Consent: The Political Economy of the Mass Media (New York: Pantheon, 1988).

[26] Daniel Dale, "Republicans release AI deepfake of James Talarico as phony videos proliferate in midterm races," CNN Politics, March 13, 2026. https://www.cnn.com/2026/03/13/politics/james-talarico-ai-deepfake-republicans-midterms

[27] Harvard Kennedy School Misinformation Review, analysis of the 2023 Slovakia parliamentary election deepfake audio.

[28] Bobby Chesney and Danielle K. Citron, "Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security," California Law Review 107, no. 6 (2019).

[29] Pope Leo XIV, presentation address for Magnifica Humanitas, Synod Hall, Vatican City, May 25, 2026. Full text and coverage: America Magazine (https://www.americamagazine.org/speeches/2026/05/25/pope-leo-xiv-calls-for-ai-to-be-disarmed-directed-to-the-common-good/), PBS NewsHour, CNN, Vatican News.

[30] Yuval Abraham, "'Lavender': The AI Machine Directing Israel's Bombing Spree in Gaza," +972 Magazine and Local Call, April 3, 2024. https://www.972mag.com/lavender-ai-israeli-army-gaza/

[31] The Guardian investigation of Israeli signals intelligence Unit 8200 storage of Palestinian communications data on Microsoft Azure, with reference to leaked Microsoft internal documents.

[32] Stuart Russell, "Take a Stand on AI Weapons," Nature, May 27, 2015 (with comment from S. Aaronson, M. Tegmark, and others); Russell, "The Third Revolution in Warfare," The Atlantic, September 11, 2021.

[33] Center for AI Safety, "Statement on AI Risk," May 30, 2023. https://aistatement.com

[34] Cade Metz, "'The Godfather of A.I.' Leaves Google and Warns of Danger Ahead," The New York Times, May 1, 2023.

[35] Water Quality Act of 1965, Public Law 89-234, 79 Stat. 903.

[36] Air Quality Act of 1967, Public Law 90-148, 81 Stat. 485 (also commonly cited as PL 90-14 in earlier draft legislative materials).

[37] TAKE IT DOWN Act, Public Law 119-12 (2025).

[38] Pope Leo XIV, "Message to Participants in the Second Annual Conference on Artificial Intelligence, Ethics, and Corporate Governance," June 17, 2025. https://www.vatican.va/content/leo-xiv/en/messages/pont-messages/2025/documents/20250617-messaggio-ia.html

[39] Hannah Arendt Center for Politics and Humanities at Bard College, "Misattributed Arendt Quotations." https://hac.bard.edu

If there are any errors in this article, please let us know.