Michael Darius Eastwood · Independent AI alignment researcher
Published
The Eden Mark is the book's name for a certification instrument that would attest to compliance with the substrate architecture at the point of manufacture. The analogy the book draws is with Fair Trade. This note describes what the Mark proposes and where the book stops.
That the chip's design was submitted to the International AI Ethics Authority for architectural review before manufacture. That the review confirmed the design includes caretaker doping at the required level and that meltdown triggers are properly integrated. That the manufactured artefact matches the reviewed design (attested through the Moral Genome Tokens). That the manufacturer has committed to ongoing audit of derived systems using the certified chip.
Fair Trade certification is a market instrument. It does not compel manufacturers to change practices; it lets consumers and enterprise buyers reward those who have. Where the certified product commands a price premium and the certification is visible, the market incentive to certify becomes real. The book argues that AI ethics compliance is a market signal buyers will increasingly ask for, especially enterprise buyers whose own liability exposure depends on the tooling they procure. The Mark is the visible signal.
Fair Trade is voluntary. The Mark, if it works, has to be more than voluntary once the field matures; the book argues for statutory review at manufacture through the HARI Treaty machinery. The Mark provides visibility; the Treaty provides enforcement. The two work together. A voluntary Mark without the Treaty would leave the incentive too weak against the compute advantages of non-compliant designs. A Treaty without the Mark would leave buyers unable to distinguish compliant from non-compliant systems in the market.
It does not certify a system's runtime behaviour. It certifies the substrate's compliance with the reviewed specification. Behaviour depends on training, deployment, and derivation; the Mark cannot follow the system into all of those. What it can do is attest that the substrate the system is running on retains the properties the Authority reviewed. That is a weaker claim than "the system is safe" and a stronger claim than "the designer said the system is safe."
The Mark is the entry ticket: a system without the Mark is not eligible for higher-envelope deployment. Graduated autonomy is the trajectory within eligibility: the envelope expands over time as the system demonstrates the orientation. The Mark is a manufacturing-side attestation; graduated autonomy is a deployment-side mechanism. Neither replaces the other.
The Mark, like the HARI Treaty and the International AI Ethics Authority, has not been adopted. The book is honest that the concrete instruments have to be drafted by people whose job that is, and that a book cannot do the drafting. What it can do is name the shape of the instruments and argue for their necessity. Six months of policy discussion since publication have not adopted the Mark by name and have not contradicted its shape.
A named certification instrument, a specific analogue drawn from an established consumer-goods market, a clear relationship with the HARI Treaty and graduated autonomy, and an honest statement about what the Mark can and cannot certify. The Mark is a market-facing instrument in a larger substrate governance architecture; it is not the architecture on its own.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.