Bermuda After AGI: A Contingency Plan
- Format
- Scenario plan, not a forecast
- Chapters
- Revenue, insurance, rule-making
- Moves
- Twelve, most can start this year
- Published
- Laurenceo, Hamilton, Bermuda
When a crisis comes, governments reach for the plan that is already written. This is that plan for Bermuda and advanced AI: three shocks it could bring to the island, and what Bermuda can decide before they arrive.
The Plan at a Glance
Laurenceo's AGI Observatory tracks when AI may be able to do the work of an AI researcher, end to end. Leading forecasts put that within a few years. Nobody knows the date. What can be known today is where Bermuda is exposed if it happens, because the exposure is already in the public accounts.
This paper does not predict when AGI arrives. It takes the scenario seriously and asks three questions. What happens to the money that runs the island. What happens to the industry that built it. And who writes the rules Bermuda will have to live under.
Illustrations, not forecasts. Each chapter sets out the exposure in numbers, then four moves Bermuda can make now. Every figure is sourced in the full plan. Scenario figures show scale and are labelled as illustrations.
01 The Revenue Hole
Bermuda has no income tax. It runs on a tax on payroll, and on the profits of the companies based here. The 2026/27 Budget projects total revenue of $2.03 billion. Corporate income tax is expected to bring in $753.2 million. Payroll tax is expected to bring in $634.9 million. Those two lines behave very differently if AI can do office work. One follows profits. The other follows paychecks.
The companies have strong reasons to stay. Bermuda's regulator, courts and speed to license are hard to copy. But a company can stay while its payroll shrinks. The table shows what that would mean, assuming payroll tax moves in step with employment income.
| Scenario | Payroll Tax Lost Each Year |
|---|---|
| A quarter of international business payroll goes | About $63 million |
| Half of international business payroll goes | About $127 million |
| All of international business payroll goes | About $254 million, more than the whole 2026/27 capital spending budget of $182 million |
Scenario figures are Laurenceo calculations: 10%, 20% and 40% of $634.9 million. They are illustrations, not forecasts.
Stress test the payroll tax base every year
The Ministry of Finance and the Fiscal Responsibility Panel should publish an annual AI exposure test: payroll tax by sector, the share that comes from roles AI can already assist, and the revenue effect of losing a tenth, a fifth and two fifths of it.
Agree an automation trigger for the fiscal rules
Write in advance what happens if payroll tax falls because of automation while corporate income tax holds or grows. Name the measure, the threshold and how much corporate income tax may then support day-to-day services. A rule agreed in calm is better than one improvised in a crisis.
Design a Bermuda Fund before it is needed
The financial assets being built from corporate income tax can become a permanent fund with a clear mandate, and a resident dividend that can be switched on by the trigger above. Alaska and Norway took years to design theirs. Starting the design costs little.
Shift the base, slowly, from jobs toward output
A tax that rises with each Bermudian hired works against Bermudian jobs in a world where the alternative is software. The 2026/27 payroll tax cuts move in the right direction. The Tax Reform Commission's work should now test each option against an AI scenario.
02 The Uninsured Risk
Every business is adopting AI. Almost none of them are insured for what it gets wrong. In June 2026 a panel at the Bermuda Captive Conference ran an informal poll. Most people in the room used AI at work and at home. None had AI liability cover. In January 2026 the body that writes standard US policy forms introduced a generative AI exclusion for general liability. Risk is being written out faster than cover is being written in.
Bermuda became the world's risk capital by building new cover after each shock: Hurricane Andrew in 1992, 11 September 2001, Hurricane Katrina in 2005. AI liability is the next one. The hard part is measurement. Whoever sets the standard for evidence will shape the market.
Make evidence the standard
The BMA's final guidance should recognise system-generated evidence: AI controls that produce their own proof as they run. That gives underwriters data they can price from. Laurenceo set this out in its consultation response, Evidence by Default.
Create an AI incident reporting standard
A simple, shared format for regulated firms to report AI failures and near misses. Cyber cover only became priceable once loss data existed. Bermuda can hold the first clean data set.
Open a clear licensing path for AI liability
Use the innovation hub, the sandbox and the Innovative Insurer class to give carriers and captives a known, fast route to write affirmative AI cover.
Model the accumulation
Publish a realistic disaster scenario for AI: one widely used model fails or is compromised and losses land across cyber, professional liability and directors' cover at once. Supervisors and reinsurers need a common yardstick.
03 The Small-State Voice
The most important decisions about AI will be made in a handful of capitals. Bermuda is not one of them. On 12 June 2026, US export controls took two of the most capable AI models on the market offline around the world. Access came back on 1 July. For 19 days, anyone who relied on those systems, in any country, went without. No small jurisdiction was consulted.
Bermuda is small, but it is not without leverage. In insurance it holds a place far above its size because large jurisdictions accept its supervision as equal to their own. That standing was built deliberately, over years, and it is the model for AI.
Use procurement as policy
Adopt standard AI terms for every public contract: audit rights, incident notice, data location, a tested exit. The Government is already deploying AI across public services with an outside partner. The contract is where a small state has real power.
Treat model access as critical infrastructure
Any essential service that depends on a frontier model, in government or in a regulated firm, should have a tested fallback to a second provider or to a manual process. June 2026 showed that access can vanish for weeks with no local say.
Form a compact of small jurisdictions
Bermuda, Cayman, Jersey, Guernsey, the Isle of Man and others face the same problem with the same limited staff. A shared position, shared model terms and a shared seat at standards bodies would carry more weight than any one of them alone.
Aim for recognition, as in insurance
Set the goal that Bermuda's supervision of AI in financial services is accepted by the EU, the UK and the US as equal to their own. That is how a small jurisdiction keeps its firms inside the room when rules tighten.
Twelve Moves and Three Signposts
Most of the twelve moves can start this year. The full plan names who leads each one and when. Three signposts say when to move faster:
- Jobs. International business payroll falls in two consecutive quarters while company numbers and profits hold.
- Losses. The first AI-caused insured loss above $100 million anywhere, or the first in a Bermuda-regulated firm at any size.
- Rules. Any large state licenses frontier models, or a second worldwide suspension of access occurs.
Open the Forecast Board Read Evidence by Default
The Full Plan
The complete 11-page plan, with the exhibits, the twelve-move table, the method, its limits and every source.
The PDF viewer works best on a larger screen. Open the plan directly instead.
Open the PDF (opens in a new tab)This plan is general information and commentary published for public discussion. It is not financial, investment, legal, tax, insurance or other professional advice. Scenario figures are illustrations built on stated assumptions, not forecasts or predictions. Laurenceo is an independent publication and is not affiliated with, endorsed by or speaking for the Government of Bermuda, the Bermuda Monetary Authority or any organisation named. If a figure is wrong, Laurenceo will correct it and say so on the corrections page.