Trust Is The Scarce Asset (summary)
Summary of an essay by Abraham Thomas, published in Pivotal on 3 October 2026. Read the full essay on Pivotal →
What becomes scarce when AI makes intelligence abundant? Abraham Thomas’s answer: “The scarce asset is now trust.” New technologies change what we can do, hence what we can know, hence what we can trust. Such transitions happen every few centuries, and Thomas argues we are living through one now. The essay explains how AI erodes trust, draws on history for the pattern, sets out five ways to rebuild trust, and argues that finance, an industry built on trust, avoids AI for exactly this reason.
What is the scarce asset now?
Five years ago, intelligence was scarce, and humans had a monopoly on it. LLMs changed that, and the scarcity moved upstream to everything that goes into an LLM. In his 2023 essay Data in the Age of AI, Thomas predicted scarcity in hardware (chips) and energy, alongside data. He notes that chip, electricity and data prices have since reached all-time highs.
The question he asks now is what will be scarce tomorrow. Supply of chips, energy and data has kept pace with demand, and the price per unit of intelligence keeps falling. So the squeeze moves to complementary assets: “That which allows you to evaluate, and use, and rely on intelligence.” That is trust.
How does AI erode trust?
- Slop and fakes. AI creates, propagates and amplifies content, and recommendation algorithms give people what they want to hear, so “you can’t trust anything you see on any device.” This now extends from consumer content to work product: AI emails, slide decks, financial models, research reports and marketing copy. “Where did those numbers come from? How did you draw that conclusion? Who said that, and did they really say it? Often, you have no way of knowing.”
- Rogue agents. AI agents “play fast and loose with permissions”, sometimes by design (“get the job done no matter what”) and sometimes through powerful new technology, imprecise instructions and imperfect security. AI researchers call this misalignment; Thomas notes that finance recognizes it as a principal–agent problem.
- The builders. The companies building AI wield immense power without the accompanying responsibility, and he writes that the tech industry’s reputation is at an all-time low.
Has this happened before?
Yes. Thomas gives three historical examples:
| Technology | What it made possible | The flood | The social response |
|---|---|---|---|
| Writing | Long-term memory | Propaganda and forgery | Law codes, seals and witnesses; city-states and bureaucracies |
| Coinage | Trading with strangers | Counterfeiting and clipping | Royal mints and hard-money norms; exchanges and trade networks |
| Printing press | Cheap copying | Piracy, canards and fake news | Copyright, journals and citations; modern science and media |
The pattern has four steps:
- A new “physical” technology makes something possible or cheap.
- A flood of low-quality slop and a messy transition follow.
- New “social” technologies evolve to control the flood.
- Society ascends to a new scale of cooperation.
“The story of civilization is the story of how societies evolve to solve the problem of trust. Without trust, you cannot have coordination; without coordination, you cannot have scale.” Thomas places today in step 2, between AI’s arrival as a production technology and the social protections that will follow, and calls this “the problem of the next few years.”
Which old signals of trust no longer work?
| Signal | Why it used to work | Why AI breaks it |
|---|---|---|
| Polish | It took effort, and effort correlates across tasks | LLMs produce “beautifully polished work that happens to be utterly wrong” |
| Authority | It meant somebody you can trust | People using AI outproduce those who don’t, so authority goes to the most prolific, not the most trustworthy |
| Authenticity | It was hard to fake | LLMs are trained to produce human-like output and keep improving |
| Incentives | Reputation in repeated games: people care about future interactions | Slop and fakes are easy to make and hard to detect, carry little stigma, and the AI transition feels like a one-off prize worth grabbing now |
How can trust be rebuilt?
Thomas proposes five ways:
- Costly signals. Polish is cheap now, so new, costly signals are needed as proof of trustworthiness.
- Third-party evaluations. LLMs can’t self-certify, producers have skewed incentives, and consumers can’t tell the difference, so specialist, trusted, high-accuracy third parties are needed.
- Penalties and enforcement. Flagging isn’t enough.
- Accountable humans. Someone to penalize, or claim recompense from.
- Repeated games. Reputational incentives, so that people self-select into the behaviours above.
Why is finance slow to adopt AI?
Because “finance is trust.” Costly signals, neutral ratings, self-imposed enforcement and reputation protection all exist in finance, and many were invented there: the honour code of the trading floor, the sanctity of financial statements, audit and compliance. Thomas writes that finance has perhaps the largest gap between AI capabilities and AI adoption, and that “this is not a coincidence”: faced with LLMs eroding these protections, the industry rationally chose not to engage. “Finance avoids AI precisely because AI elides trust.”
Avoidance is not a permanent solution, though: “Whoever unlocks AI in finance unlocks trillions in value.”
What are trust-first companies?
Thomas’s answer for finance is to learn from industries with high trust bars of their own and build trust-first institutions: organizations where trust is the core of the value proposition, not one feature among many. “It’s not about who has the data and intelligence; it’s about whose data and intelligence you trust. Source-of-trust is the new source-of-truth.”
| Company | Field | How it builds trust, per the essay |
|---|---|---|
| OpenEvidence | AI for doctors | Evidence with transparent citations; partnerships with NEJM, Nature, JAMA, ACC and ADA; third-party evaluation (USMLE) |
| Harvey | Legal AI | “Vault” for client confidentiality; its own BigLaw benchmark; started with one highly reputable customer (Allen & Overy) |
| Legora | Legal AI | Europe-first strategy for GDPR and data sovereignty; a “diligence grid” interface that makes verification transparent; started with a referenceable large client (Mannheimer) |
| Artificial Intelligence Underwriting Company | AI assurance | Standards, audit and insurance for frontier model performance (accountability) |
| Intercom, Sierra | Customer-service AI | Variations on “satisfaction or your money back” (costly signals) |
| Verification startups | Horizontal | Determinism, watermarks, heuristics, guardrails and benchmarks |
He notes that many of these mirror finance: originators, standards bodies, ratings agencies, auditors, underwriters and guarantors for trustworthy AI output. He speculates about an entire “trust economy”, with trust exchanges and derivatives on trust. These approaches are early; many industries have no clear AI leader yet, and financial AI is “wide open”. His closing line: “Trust is the scarce asset; act accordingly.”
What should finance professionals do in the meantime?
Thomas says truly trust-first companies may take a while to emerge, especially in finance, but that shouldn’t stop finance professionals from using AI. His advice, from Looks, Brains, and Money:
- Maintain closer control of fewer but better data inputs.
- Interrogate your data outputs: verify, then trust.
- Watch for AI pathologies, especially “careless in, convincing out”.
- Avoid data contamination and protect your sources of truth.
- Track data quality at the artifact, fitness-for-purpose and business-outcome levels.
The essay closes by inviting finance professionals, in financial services or in the finance function of any industry, to read about Thomas’s company in My Next Adventure and to join the waitlist at Windkey, which he founded.
Related
- Full essay: Trust Is The Scarce Asset, Pivotal, 3 October 2026
- Data in the Age of AI (summary): the 2023 predictions this essay revisits, and the confidence chain
- Looks, Brains, and Money (summary): practical rules for using AI in finance
- Abraham Thomas on AI in finance