Lists of the most powerful people in A.I. tend to measure the same things: who owns the models, who owns the chips, who signs the largest checks and who can build the future. The same names recur: the builders, the chipmakers, the financiers and increasingly, the researchers shaping the technology itself. Those rankings capture an important dimension of power. But they also raise a deeper question: as A.I. becomes more widely available, what forms of influence become more valuable, not less?
A pattern repeats often enough to look like a law: when technology makes something abundant, power rarely disappears. It relocates to whatever becomes scarce. The logic of power has always followed ownership of the scarce input: land when kings held it, oil when the barons did and computing power now that the machines need it.
Yet the irony is that even one of the most famous barons of them all was selling trust. John D. Rockefeller called his company Standard Oil because, as Ron Chernow records, kerosene at the time was so unevenly refined that it killed thousands of people a year in lamp explosions. The name “Standard” was itself a promise: that the product could be trusted.
Economist Herbert Simon supplied the modern version of the same idea in 1971: “a wealth of information creates a poverty of attention.” The question worth asking about A.I., then, is what it makes abundant, and what that abundance makes scarce.
What A.I. makes abundant is intelligence, or at least a persuasive imitation of it. The cost of running a model at a fixed level of capability fell roughly 280-fold in roughly two years. Whatever the frontier laboratories ship, an open-weight equivalent now arrives about four months later. The frontier still matters. By Stanford’s AI Index, the best closed models continue to outperform the best open-weight models. However, a lead measured in months is a lease, not an estate.
And what abundant intelligence produces is plausible competence. Models can generate possibilities almost without limit. They cannot accept responsibility for choosing between those possibilities. That is where judgment becomes scarce. A public database now tracks more than 1,700 court decisions involving fabricated A.I. citations. Even the statistics describing the flood have become part of the flood. One widely quoted claim, that 90 percent of online content will soon be machine-generated, traces back to a 2020 trade book. It was footnoted in a Europol report and then repeated as research for years. The 2024 revision removed the earlier reference, but by then the claim had acquired a life of its own, cited across the internet as part of a wider algorithmic drift.
A market in proof is already under construction. Cloudflare, which sits in front of roughly a fifth of the web, changed its default in 2025 to block A.I. crawlers from ingesting human-made pages unless they pay. Google reportedly pays Reddit $60 million a year for the right to train on human conversation. These are markets in provenance, the bare fact of human origin.
But provenance is only the beginning. Knowing that a human made something says nothing about whether it is true, or safe to act on. The deeper scarcity is judgment: the capacity to discern and decide, and to be believed.
Television got there before the analysts. In Game of Thrones, the spymaster Varys poses a riddle. A king, a priest and a rich man sit in a room with a common “sellsword,” and each commands him to kill the other two. Who lives and who dies? “Power resides where men believe it resides,” Varys answers. His answer points to an emerging group: the judgment class.
The A.I. economy has created its own sellsword: a mercenary capable of producing almost anything for anyone at a price per token. The old riddle turned on what the sellsword believed; this one believes nothing and serves everyone, so the deciding belief moves to whoever acts on its output. The power lists rank the crown and the coin, but the riddle says they were never the point.
A study by KPMG and the University of Melbourne, spanning 47 countries, found that fewer than half of people trusted A.I., while almost two-thirds admitted relying on its output without checking it. That gap matters because trust behaves like a form of capital. It is accumulated slowly and lost quickly, and it does not come back at the old price. Deloitte refunded part of a government contract in Australia after an A.I.-assisted report was found to contain invented sources. Follow where trust is being bought and sold, and a different set of gatekeepers comes into view.
The Big Four accounting firms are building A.I. assurance into a paid service line, selling the right to say that a system has been tested. Underwriters at Lloyd’s of London have begun writing policies against A.I. hallucination, with models assessed before cover is issued and payouts triggered if agreed performance thresholds are no longer met. Lloyd’s priced seaworthiness long before governments regulated it. Now it is helping to price acceptable machine error.
Universities are returning to handwritten, proctored exams because the take-home essay no longer proves what it once did. Sweden is spending more than $100 million to put printed textbooks back into its schools, after concerns that screen-first classrooms had come at the expense of attention and deep reading, and Norway has gone further, announcing a near-ban on A.I. in its elementary schools. Different institutions are responding in different ways, but all are retreating to systems they can trust.
The obvious objection is that the model owners are not becoming commodities at all. They are becoming more powerful. The hyperscalers plan roughly $725 billion in capital spending in 2026, and the companies building frontier models remain concentrated in relatively few hands. That only sharpens the point.
A handful of firms hold durable infrastructure power. The race for the crown is crowded, international and largely invisible. But for everyone downstream of them, which is almost every company, board and profession, models are becoming an accessible input, priced by the token.
That helps explain why adoption has run so far ahead of readiness, even in organizations without the resources to deploy the technology well. In many executive teams, the motive is not a bid for dominance. It is the fear of losing power: of being the last one holding an asset that has stopped being scarce.
Most power lists naturally emphasize the people building and financing A.I. But there is another form of influence emerging alongside them: the people and institutions whose judgment determines whether A.I.’s outputs are trusted and acted upon.
The judgment class does not photograph well. It includes the auditor whose signature still carries weight, the editor whose byline functions as a warranty, the judge, the examiner, the underwriter, the scientist whose work survives scrutiny and the executive whose sign-off nobody feels compelled to re-verify.
What they possess is the one asset A.I. makes more valuable as it becomes more capable, because every improvement in the machine’s fluency raises the price of knowing when to believe it.
The test for anyone holding power today is simple. Ask what that position actually rests on. If the answer is access, information, capital or compute, it rests on something technology is making steadily easier to obtain. If the answer is that people act on your judgment without feeling the need to check it, you possess something much scarcer.
The next chapter in A.I.’s power story may belong to the judgment class. Most of its members would never think to include themselves.
Rahim Hirji is a future-of-work strategist, founder of The SuperSkills Intelligence Company and author of SuperSkills: The seven human skills for the age of AI, published by Kogan Page and out now.
