The Hallucination Economy
How organisations act on AI claims no one can check, and what boards should ask instead
The evidence comes from the vendor, from the system itself or from someone with a stake in the answer. Decisions rest on it anyway.
Claims nobody else can test
In February 2026 the Reserve Bank of New Zealand asked ten AI models to help measure how exposed the country's jobs are to AI, and said in its own findings where the models' answers varied most.
Most claims about AI arrive without that. The evidence comes from the vendor, from the system itself or from someone with a stake in the answer, and nobody else can test it. Decisions about money, work and public services rest on it anyway. That is the hallucination economy.
I follow those claims through five stages: what organisations say about their AI, the records beneath those statements, the platforms that hold the records, the regulators meant to referee them, and the check a board can run. I went back to my own published figures too, and not all of them survived.
Written from Aotearoa New Zealand, a country that buys the machinery and builds none of it.
What the book gives you
- Seven properties that make a claim about an AI system checkable by someone other than the claimant
- The questions a board can put on Monday morning, and what counts as an answer
- Worked cases from vendor filings, benchmarks, surveys and regulators in the US, Europe, Australia and Aotearoa
- A record of how the book was drafted with AI and checked, with the method published so you can re-run it
From the book
The Hamberger Law of AI AdoptionAdoption velocity follows a doubling curve; governance maturity grows in linear increments. The gap between them is not a lag. It is a structural condition that widens faster the more successful the adoption becomes.
Chapter 19 · The Board's Verification QuestionsThe general principle is older. Larry Downes and Chunka Mui stated it as the Law of Disruption in 1998, and Scott Brinker as Martec's Law in 2013. The Hamberger Law of AI Adoption is its specialisation to AI governance, with a testable rate claim.
Every substantive figure in this chapter was produced by a party with a position in the answer.
Chapter 2 · The 81 Per Cent Problem
Nobody in the chain is required to say how it was established.
Chapter 1 · Adoption Theatre
Inside: nineteen chapters
- Adoption Theatre
- The 81 Per Cent Problem
- A Leadership Problem, Not a Skills Problem
- The Cognitive Ceiling
- Unverified Chains
- The Benchmark You Cannot Audit
- The Wall That Was Never There
- The Record Was the Target
- Who Wrote This Code
- The Control Plane Ate the Model
- A Router and a Front Door
- The Evaluator You Share
- The Brussels Sequence
- When AI Became State Infrastructure
- Digital Sovereignty
- The Constitutional Standoff
- The Anatomy of a Checkable Claim
- What a Country That Owns None of the Machinery Can Check
- The Board's Verification Questions
About the author
Andreas Hamberger is a New Zealand technology leader, enterprise architect and board director. His work spans enterprise architecture, AI governance and open source. He holds a Master of Arts in the Philosophy of Logic and founded Te Pono Limited.
Also by Andreas Hamberger: Enterprise Architecture in Practice, Space Mafia, Free as in Theft, Lethal by Design, The Manifesto of the Humanist Party.