Who Pays for What We Know

Bernardine Cantwell/Entrance to a brick building labeled "Carnegie Free Library"

Artificial intelligence is producing fortunes on a Gilded Age scale, and a familiar pattern with them. The firms building the technology already produce most of what is known about it. According to Stanford’s AI Index, over ninety percent of notable AI models are produced by private industry. In the U.S. and Britain, government bodies test frontier models before release—but the testing is done under voluntary agreements that developers can decline to renew.

Now those firms and their founders are also becoming leading funders of research on AI’s consequences, through a wave of philanthropy without precedent since Carnegie and Rockefeller. The firms whose products are the subject of inquiry, and the individuals whose fortunes depend on them, are becoming leading financiers of research on AI’s consequences. 

The OpenAI Foundation, the nonprofit that controls OpenAI, emerged from the company’s 2025 restructuring holding roughly a quarter of the company. That October it committed $25 billion to curing diseases and what it calls “AI resilience,” and in March 2026 it pledged to grant at least $1 billion over the following year, including a resilience program led by an OpenAI cofounder Wojciech Zaremba.

Anthropic’s founders have made a commitment of comparable scale. In January 2026, CEO Dario Amodei wrote that those growing rich from AI bear a responsibility to address its economic consequences. The company’s IPO filing, made public this October, confirmed that each of its seven cofounders has pledged eighty percent of their personal equity to philanthropy, a sum estimated at nearly $80 billion. The company has also placed five percent of its shares in a new Anthropic Public Benefit Fund and pledged $200 million to an Economic Futures Research Fund for outside research on AI’s economic effects.

The chief executive of Coefficient Giving, formerly Open Philanthropy, has estimated that AI wealth could add some $40 billion a year to American philanthropy if Anthropic and OpenAI go public.

As the builders of artificial intelligence become the leading philanthropists of our era, which questions attract attention and talent will be increasingly shaped by a rarefied Bay Area clique. What questions will go neglected as a result, and what institutions can hope to check their newfound power?

Industrialist Charity and Industry Power


In May 1913, the richest man in the world donated $100 million dollars—about a ninth of the fortune he had built by revolutionizing how the world was fueled—to fund a new foundation intended “to promote the well-being of mankind throughout the world.” It bore his name, and his son, John D. Rockefeller, Jr., was its first president. The organizing principle of the Rockefeller Foundation was “scientific philanthropy”: striving to systematically “cure evils at their source.”

In September 1913, more than 10,000 coal miners in southern Colorado went on strike to demand better working conditions and the right to live and shop outside the mining companies’ control. Evicted from company housing, they set up tent colonies at the mouths of the mines to house strikers and to block replacement workers. Unrest would last until the spring.

On April 20th 1914, the Colorado National Guard and company guards opened fire on the colony at Ludlow and set its tents ablaze. About twenty people died that day, among them two women and eleven children who suffocated or burned to death. The largest operator in the Colorado coalfields was Colorado Fuel and Iron, a Rockefeller company.

Two months later, the Rockefeller Foundation created the Department of Industrial Research. The new department was charged with investigating and offering “constructive suggestions concerning industrial and social policies.” Its director, William Lyon Mackenzie King, was also John D. Rockefeller Jr.’s personal advisor on the Colorado strike. King’s counsel produced the “Rockefeller Plan,” adopted at Colorado Fuel and Iron in 1915, which gave miners a company-sponsored channel for grievances, though not the independent union they had struck for. The plan would become a template for company unions around the world.

When the Commission on Industrial Relations investigated Ludlow, it took a critical view of the arrangement. Congress had created the commission in 1912 to explain why American labor relations had become so violent and to develop recommendations for the government. Its chairman, Frank Walsh, made the Rockefeller Foundation itself a subject of inquiry. Walsh argued that research paid for by industry could further entrench the power of that industry by shaping public debate and belief. The Rockefeller Plan illustrates this point.

More than a century later, technology is again revolutionizing our economy. New fortunes are being built on this economic revolution. Effective Altruism parallels scientific philanthropy, and Walsh’s critique is relevant again.

Philanthropic organizations now play a central role in American political life, shaping which problems are salient to the public and what actions should be taken by government, companies, and individuals. Like the institutions Walsh scrutinized, they still depend on patronage, and those who benefit most from the existing economic and political orders are often those best equipped to finance its study. This no doubt biases the sorts of ideas that get produced; however, the biggest source of bias of all arguably lies with the dog that didn’t bark: the questions and fields that receive no philanthropic support whatsoever.

Patronage’s greatest influence lies in its ability to direct our attention. We don’t have to be cynical about funders or researchers to worry about this—we can believe that they are sincere people doing good work, while still noticing how a society’s understanding of its economic order is disproportionately shaped by those that the economic order has rewarded. That influence can coexist with careful, sincere, admirable work, and often does.

 But it means a society’s understanding of its own economic order is shaped, in part, by those that same order has rewarded. The Progressive Era, in contrast, built public institutions that could ask questions of their own, offsetting the power of the Rockefellers of the world. 

Gilded Age fortunes financed a vast expansion of American intellectual life, from Carnegie’s libraries to the elder Rockefeller’s University of Chicago and Institute for Medical Research. The idea had been planted a decade earlier by Andrew Carnegie’s 1889 essay “Wealth,” later known as “The Gospel of Wealth.” Carnegie argued that the wealthy had a moral obligation to administer their surplus during their lifetime to help the poor improve themselves. “The man who dies thus rich,” he wrote, “dies disgraced.” Rockefeller read the essay and wrote to Carnegie in admiration. 

This way of spending was definitely better for the world than buying yachts and houses like the Goulds and Commodore Vanderbilt, and the institutions it produced did enormous good. But the “Gospel of Wealth” assumed that people who are good at ruthlessly looking out for their own interests in business will easily be able to switch to looking out for the interests of others in philanthropy. Donors didn’t have to directly dictate conclusions to shape the limits of idea-space. The limits come from their choices of what to fund, and on what terms.

In 1881, Joseph Wharton, a co-founder of Bethlehem Steel, donated $100,000 to the University of Pennsylvania to fund a new School of Finance and Economy. Now known as Wharton, it was the very first collegiate business school. Early research at Wharton includes Simon Patten’s The Economic Basis of Protection (1890), which argued against international free trade in favor of government protections of domestic industry. Wharton also fostered George W. Taylor, who invented the field of industrial relations, pioneering practical work in how labor and management could solve disputes without striking. On the other hand, Wharton declined to renew the professorship of Scott Nearing, who advocated against child labor.

Turning Knowledge Into Power

Amodei’s argument today is, in substance, Carnegie’s, and there is as much to admire in it now as there was in 1889. Much of this money will likely go to medicine and other causes unrelated to AI. But much of it will go towards researching the effects of AI on workers, children, security, and the economy—and society’s possible responses. Some theories will be better than others for the AI Labs’ bottom lines and the AI researcher’s personal and class self-regard.

Profit is not the only all too human motive. When Frederich Engles managed his family’s textile and cotton-spinning business, he was earning to give to his then underemployed friend and philosopher Karl Marx. He was not legitimizing capitalism but was absolving his own role as a capitalist, funding the good (intellectual) fight.

The current technological revolution intensifies these questions. Rockefeller’s power was over energy and the money energy could buy. The AI tycoon’s power is over information itself, as decision-makers and the general public come to rely on AI systems to delegate away laborious thinking to the AI Labs themselves. Even individuals and organizations that seem from the outside independent might internally be using AI to manufacture their output and even their views. 

The British-Hungarian polymath Michael Polanyi’s account of the “republic of science” explains how discovery depends on the freedom of investigators to pursue unexpected questions. A researcher could be totally free to publish an unwelcome conclusion while having no realistic way to pursue a question that lies outside established funding priorities. Formal intellectual freedom and practical intellectual capacity are not the same thing.

A problem has to be described before it can be solved. What’s causing the problem? How bad is it? How do we even measure how bad it is? The answers shape the space of possible action in response.

Elected policymakers aren’t researchers. They want to do something about problems—which means that they mostly tackle problems that have been measured, modeled, and translated into proposals. Questions without established methods or advocates may never enter political consideration at all. We’ve seen how this played out with the ideological slant of early research at Wharton.

The Big Lab governance arrangements recall the Rockefeller charter fight. The founders’ pledges are, in the filing’s own words, voluntary and revocable, and Anthropic’s cofounders will retain majority voting control of the company even as they give their economic stakes away. The OpenAI Foundation is a public charity; unlike private foundations, it does not have a twenty percent cap on business holdings and mandatory five percent annual payouts of investment assets. Observers have wondered how separate the foundation really is from the company it controls

Some funders have tried to address the conflict of interests directly. Dustin Moskovitz and Cari Tuna, Coefficient Giving’s principal donors and early investors in Anthropic, moved their stake into a nonprofit vehicle in 2025 in part to dispel any perception of a conflict of interest. The labs describe their grants as support for independent work. There’s no reason to doubt their sincerity. But we don’t need to be cynical about intentions to acknowledge that when a large share of the money for studying a technology comes from those who profit from that technology, research may be biased.

Let’s distinguish between institutional capture and institutional dependence. Institutional capture occurs when an institution charged with serving the public is induced to serve a private interest instead. Dependence is not as malicious. A researcher pursues the truth conscientiously—within an agenda shaped by a few funders. A legislature has the power to set rules for a technology—but no independent means of understanding it. Ethics rules address capture, but not the deeper question of who has the capacity to produce knowledge in the first place.

Public funding is not automatically pure either. Governments have their own incentives. Private philanthropy can sustain intellectual diversity that would be impossible in a totalitarian state. What matters most is whether an intellectual system can sustain competing lines of inquiry, or whether the system depends on concentrated sources of support that narrow the range of possible questions it can pursue. Government and industry are both biased, but ideally these competing incentives serve as mutual checks and balances.

What recurs is the underlying political problem Walsh confronted: economic transformation changes the distribution of epistemic capacity faster than political institutions can develop the means to understand it.

The Progressive Answer Then and Now

The rise of the great foundations was a step in genuine moral progress compared to the Jay Gould and Commodore Vanderbilt types who had come before. But it was also an atavistic return to aristocracy. Eighty years before, the French diplomat and historian Alexis de Tocqueville had, in Democracy in America, praised that young country’s system of voluntary associations. In an aristocratic society, a few powerful men could undertake great works on their own. In a democracy, Tocqueville said, free associations of equal citizens took the place of the powerful private persons whom equality had swept away. The modern foundation substantially reversed that Tocquevillian substitution.

Tocqueville had seen this coming. He warned that industry might breed a new aristocracy of manufacturers—predicting it would be one of the harshest the world had ever seen, but also narrower and less dangerous than the ancient regime, because its members shared no common spirit and sought no role in governing. Still, he wrote, the friends of democracy should “keep their eyes anxiously fixed” on it. The new aristocracy of wealth had lacked a role in public life. The private foundation changed that. Through private foundations, industrial wealth could act on public life—not as an interest to be checked, but as a patron of the common good.

The Gilded Age was also the Progressive Era. Governments built long-term capacities for understanding the economy they governed: the Bureau of Labor Statistics, the Federal Trade Commission, the agricultural experiment stations of the land-grant universities, and state labor bureaus. We see how this played out in the Ludlow controversy. The Commission on Industrial Relations compelled testimony from workers, managers, and the heads of the largest fortunes in America. The record of industrial conditions that the committee assembled was not the kind of thing that any privately funded body was likely to produce.

Research funded by private philanthropy continued to shape American life—often for better. But private patronage no longer held a monopoly on the knowledge the public needed.

That public capacity has since eroded. Between 1979 and 2015, the staff of the Government Accountability Office has shrunk by 44 percent, the Congressional Research Service by 28 percent, and congressional committee staff by 38 percent. In 1995, Congress eliminated its own source of independent technical analysis, the Office of Technology Assessment. But nature abhors a vacuum. The political scientists Timothy LaPira and Herschel Thomas found that more than half of registered federal lobbyists are former government officials, and LaPira describes them as filling “a vacuum in Congress” for expertise.

The more expertise is available outside government—from lobbyists, think tanks, and philanthropically funded institutes—the easier it becomes to forgo paying for expertise inside the government. The less expertise the government keeps, the more it must borrow. Outside institutions also compete with government for the same scarce specialists, usually at higher salaries. Dependence, once established, tends to deepen.

There is no way around it: when a new technology changes an economy beyond comprehension, the research to attempt to comprehend the economy again will always be primarily funded by the people that the economy rewarded. Although it can’t be eliminated, the structure of dependence can be examined—and when examined, hopefully mitigated.

Progressive Era politicians built institutions to understand the new economy they governed—independent of the powerful interests they were governing. They knew how much legitimacy depends on epistemics. But they found out late, in response to massacres and unrest, and they never fully solved the problem of epistemic independence. Frank Walsh could force one of the famously private Rockefellers to testify, but he couldn’t decide which questions the nation’s experts would spend the following decades investigating. That power belonged to those who paid for the research.

It still does.

Grace Werner is the VP of Growth at the Foundation for American Innovation. You can follow her at @poastmilque.