The Actuarial Sovereign
Nobody from the government visits the company in this story. It gets governed anyway — by a renewal letter, a procurement checklist and a terms-of-service update.
This story is from the AI Scenario Explorer, which pairs eight forces shaping AI, two at a time, into 28 maps — each map with four corners, each corner a different way that pairing could play out. A vignette takes one corner and makes it concrete: structured conjecture, not prediction. About the project →
Nobody from the government visits the company in this story. It gets governed anyway — tightly, within a year — by three documents, none of which is law.
The company is mid-sized and unglamorous. It handles warranty claims for appliance manufacturers, and in March it switches on a system that reads each claim, checks the entitlement and settles the routine ones end to end. In its first week: a failed dishwasher pump, €118, approved and paid in nine seconds. A person sees only the exceptions.
What does the law say about this? A two-year-old consultation paper, a bill in committee, guidance that is advisory, three jurisdictions with three draft regimes that disagree. Waiting for public rules to settle would mean waiting years. Like everyone else, the company does not wait.
Then the renewal letter arrives.
The company's liability insurer has added an AI endorsement. Cover continues on conditions: every automated settlement logged with the model version that produced it; amounts above a threshold routed to a person; the model pinned to versions the insurer has evaluated; incidents reported within seventy-two hours; a daily cap on what may be settled without a human. The letter is not law. But breach a condition and the company is uninsured — one bad quarter from insolvency. The conditions go up on the claims-office wall, and compliance quietly reorganises around them.
The conditions travel, too. In the autumn the company's largest client adds a mandatory field to its procurement portal: evidence of AI liability cover and the controls behind it. The insurer's clauses have become the supply chain's entry ticket. A competitor without cover does not lose in court; it fails a checklist, misses a shortlist, and never finds out why.
The model vendor is a different kind of power. Its updated terms reserve the right to audit and to suspend the endpoint, and one Tuesday a capability the claims system relies on is switched off upstream — globally, overnight — over something a different customer did in a different industry. No notice period a parliament would recognise; no one to appeal to. The three draft regimes disagree with each other; the vendor's terms are identical in every country the company operates in. The only harmonised law of AI in this story is a contract.
By December the claims system is governed — genuinely governed. Logged, capped, version-pinned, escalated, reported, attested: a tighter regime than the draft bill would have imposed, enforced faster than any court, because the sanctions are automatic. The cover lapses; the contract fails to renew; the endpoint goes dark.
The rules were written by underwriters minimising a loss ratio, a procurement department minimising supplier risk, a trust-and-safety team minimising platform liability. None elected, none obliged to consult, none accountable to the people the system decides about — and in all their careful clauses, the claimant is never mentioned once. A harm that is diffuse, or slow, or falls on someone uninsured never moves a loss ratio, and goes ungoverned entirely. A regulation can be read, challenged and appealed by anyone it touches; an underwriting model cannot.
This is a vignette from the AI Scenario Explorer — structured conjecture, not reportage: one corner of Institutional adaptation × Governance coordination given room to breathe. Signs it is arriving: named AI cover or explicit AI exclusions as standard policy terms; a deployment killed or reshaped by insurability rather than by a regulator. Signs it is not: statute binding first; assurance cheap enough that buyers verify systems directly; insurers refusing to price the risk at all — a wall, not a rulebook, and a wall does not govern so much as relocate.
When public rulemaking is too slow and too fragmented to bind, governance does not stop. It migrates to whoever prices the risk — and to whoever can switch the system off. Whether that is stopgap or settlement depends on whether public institutions catch up, and on who gets a say, in the meantime, in a rulebook nobody voted for.