Epistemic Debt: The Economics of Ungoverned AI

Ungoverned AI systems that lack verification capability create uninsurable structural liability. They do not fail randomly. They systematically convert epistemic risk into financial, legal and institutional liability.

This condition is not inherent to AI use: governed deployment restores attribution, bounds liability and re-establishes insurability by reinstating external verification authority.

This paper introduces Epistemic Debt as a balance-sheet construct: the accumulated liability incurred when consequential decisions are made without verification while organisational verification capacity simultaneously erodes. Unlike technical debt, epistemic debt compounds silently, becomes visible only at the point of failure and may be irrecoverable once an organisation no longer retains the expertise required to audit its own decisions.

Building on prior architectural and evaluative results, the paper maps failure propagation from individual confabulation through functional contamination to institution-level and market-level exposure. It demonstrates why V = 0 systems violate the conditions of insurability, why auditor and fiduciary liability are materially altered and why strategic convergence on identical unverified substrates creates correlated fragility across sectors.

The conclusion is binary choice. Organisations either invest in governance early and visibly or pay for failure late and invisibly. There is no middle path.

Papers One to Three complete the architectural analysis on which this paper depends. This paper translates those structural constraints into institutional liability. Paper Five extends that translation into market mispricing. The governance requirement established earlier in the series is not discretionary. It is structural compensation for an absent capability.