Moral Genome Tokens: cryptographic verification of the substrate architecture

4 min read · 753 words
Share:
Michael Darius Eastwood
Michael Darius Eastwood · Independent AI alignment researcher
Published
Michael Darius Eastwood · From the book · 3 July 2026
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256 anchored: f0d1f38f).

Moral Genome Tokens are the book's proposed cryptographic mechanism for verifying that a system was built to the caretaker doping specification. This note explains what the tokens do and what they do not.

"Moral Genome Tokens: cryptographic signatures embedded throughout the system that verify the presence and integrity of the ethical architecture. Just as biological DNA encodes the information needed to build an organism, Moral Genome Tokens encode the information needed to verify that an AI system was built according to Eden Protocol specifications."

What the tokens do

They provide cryptographic evidence that specific components of the ethical architecture were built in a specific way and have not been altered. A third party can verify the tokens without full access to the system's internal state; the verification is a check against a signed manifest of what the system is supposed to contain. This is the same class of instrument as code-signing certificates, hardware attestation tokens, and supply-chain attestations, applied to the substrate architecture.

Why cryptographic and not procedural

Procedural verification (an inspector visits the lab, checks the design, signs a piece of paper) does not survive derivation, fine-tuning, or distillation. A system that was procedurally verified at one point can be derived into a system that is not procedurally verifiable at a later point without any of the paper trail changing. Cryptographic verification travels with the artefact. If the tokens are present and unaltered, the specific architectural claim they attest to is verifiable; if the tokens are absent or altered, the artefact fails verification.

What the tokens do not do

They do not prove the architecture is safe. They prove the architecture is what the manifest says it is. If the manifest is wrong (either because the specification was wrong or because the implementer misunderstood it), the tokens attest to something that is not what the reader thought was being attested to. The tokens are a supply-chain instrument; they are not a safety proof.

How they fit with the HARI Treaty

The HARI Treaty as the book proposes it requires review of chip designs at manufacture. The tokens are the ongoing verification instrument for what was reviewed: they let a downstream integrator confirm that the delivered artefact matches the reviewed design. Without the tokens, the review is a one-time event whose validity decays as the artefact moves through the supply chain. With the tokens, the review is a persisting property.

Why the biology metaphor

Because DNA is the closest natural analogue to the intended mechanism. DNA does not prove an organism is healthy; it encodes what the organism is supposed to be, and comparison against the encoding is how variance is detected. That is the operational role the tokens play. The book is careful that the metaphor is at that operational level, not at the level of claiming AI systems have biology.

What the field would need to build to instantiate this

A specification of what the tokens attest to (which requires the underlying substrate architecture to be specified enough to sign). A trust root for the signing (which requires the International AI Ethics Authority or a functional equivalent). A verification tooling ecosystem (which requires the token format to be public and stable). None of these exist in operational form as of the book's publication or as of mid-2026. The book flags them as necessary work.

What the reader keeps

A named cryptographic instrument, a supply-chain role for the instrument, a candid distinction between attestation and safety, and a candid list of what has to be built for the instrument to be usable. Not a completed system. A specification for a class of instrument the field will need.

From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.

Buy on Amazon UK Amazon US

Stay informed

New posts on AI alignment, convergence evidence, and the ARC/Eden research programme.

Get updates →