The tax that protected truth

For centuries, what protected us from lies wasn’t virtue. It was cost. AI has broken that equation. What happens to truth when verification costs a thousand times more than fabrication?

Faking a study, fabricating evidence, sustaining an elaborate deception used to cost time, talent, and money. We never called that “protection” because it was invisible: the high cost of producing falsehood acted as a filter that defended us without our knowing it.

Peer review was built on the assumption that writing a scientific paper took months of work. That is why a small number of volunteer reviewers could carry the load: in 2020 alone, academics spent more than 100 million unpaid hours reviewing manuscripts ¹. Today, more than 3.2 million scientific articles are published every year, 48% more than in 2015 ¹, and Retraction Watch has now documented more than 65,000 retractions ², some containing phrases as revealing as “As an AI language model, I…” that slipped past expert reviewers ⁷.

Fact-checking faces the same mismatch: 443 organizations worldwide produced 38,000 fact checks in the first five months of 2025, 6% fewer than the year before ³, while AI-generated content had already matched and even surpassed human-written content by volume on the internet in 2024 ⁴.

And in the courts, a British lawyer had to call in forensic experts to prove that threatening audio in a custody case was a deepfake made with software and tutorials available online ⁵. None of these institutions is failing out of laziness. Their entire architecture was calibrated for a world where output was scarce. AI doesn’t break a specific rule: it pulls the ground out from under them.

This is where Brandolini’s law comes in: refuting nonsense takes an order of magnitude more effort than producing it ⁶. It isn’t new; it’s as old as propaganda. What has changed isn’t the asymmetry, but the scale. For centuries, we tolerated that tax because lying was costly too: whoever fabricated a deception paid a toll similar to the one paid by whoever dismantled it. AI breaks that symmetry of costs: it reduces the price of producing the plausible to zero while leaving the price of verification untouched. The reviewer, the journalist, and the judge are now fighting a fire hose with a watering can. We haven’t lost the truth. We’ve lost the tax that protected it.

Our verification institutions don’t need more resources. They need a design that doesn’t depend on scarcity. And we don’t have one yet. If the friction that protected us has disappeared and we still haven’t built what will replace it, what do we rely on in the meantime to decide what is true?