Concept · first placed on the public record 2026-01-02
Moral Assurance Bonds
A financial-instrument proposal requiring AI developers to post bonds against potential downstream harm, creating a direct economic incentive for genuine safety work. Policy proposal.
Moral Assurance Bonds are the financial-instrument layer of the policy stack. The proposal is that developers of frontier AI systems post bonds against downstream harm as a condition of deployment. The bond is drawn if specified harms occur. The bond is returned if they do not. The purpose is to convert an externality into a priced position on the developer's balance sheet.
Why the mechanism is proposed. Ordinary product-liability regimes cover harms after the fact and cover them unevenly across borders. Bonds cover the same territory in a different currency: they price the risk before the harm occurs and give the developer a marketable position that reflects that price. When alignment standards are met, the price of the bond falls; when they are not, the price rises; when a specified harm occurs, the bond is drawn. That is a familiar structure from other high-consequence domains and it is proposed here as a fit for frontier AI.
How it fits with the rest of the stack. Bonds pair with the Eden Mark Certification (which gives buyers a signal about which products are eligible), with the HARI Treaty (which defines the harms the bond is drawn against), and with the IAEAI (which arbitrates disputes about whether a specified harm has occurred). Together the pieces make an incentive stack that is coherent from manufacturer to buyer to state.
What the bond is not. It is not a licence fee. It is not a fine. It is not a cap on liability. The bond is a returnable position; if the standards are met and no drawing event occurs, the bond returns to the poster. That structural feature is what distinguishes it from a punitive instrument and what makes it usable as an ordinary market position.
How it interacts with the alignment argument. The book's overall claim is that alignment must be embedded rather than filtered. Bonds do not settle that architectural question; they settle a separate incentive question. Given that architectural alignment is the aim, bonds give developers an economic reason to pay for the deeper work rather than the surface work. They convert a research virtue into a cash-flow advantage.
Related: HARI Treaty; Eden Mark Certification; IAEAI; Semiconductor Chokepoint; Verification Challenge.
Evidence anchors. First placed on the public record 2 January 2026 in Infinite Architects (ISBN 978-1806056200).
Related concepts
Sources and evidence
- Primary source: https://www.michaeldariuseastwood.com/priority-claims
- Priority and Claims Ledger: /priority-claims
- Start here: /start-here.html
- Machine-readable dataset: /research/data/concepts-pages.json
- How to cite the whole programme: /how-to-cite/
Priority is established by the date of publication, not by credential or affiliation. This concept was first placed on the public record on 2026-01-02 (2026). Priority applies to the specific named framing under this label; where adjacent ideas have established prior art, no invention claim is made over it. See the corresponding entry in the Priority and Claims Ledger for the anchored first-publication line and any concept-specific risk notes.
Generated by scripts/build-concepts-pages.mjs from research/data/concepts-pages.json. British English throughout. Deterministic; contains no runtime timestamps.