Michael Darius Eastwood · Independent AI alignment researcher
Published
The manuscript that anchors the register was self-emailed at 02:45:18 UTC on 8 December 2024. Five attachments, 221,236 words, one.eml file whose SHA-256 hash begins f0d1f38f. Anyone can download it. This article explains what it says, so readers do not have to take the summary on trust.
the sealed source .eml (held privately, available on request, hash-anchored; published SHA-256 f0d1f38f…) · /priority-claims · /research/priority.json.At HRIH line 14005, one sentence: "In the long run, AI systems cannot be truly controlled. They will evolve beyond any safeguards we put in place." This is the passage the register measures against Anthropic's 18 December 2024 alignment-faking paper. It is a directional claim, not a numerical prediction. The register credits Anthropic with the experiment, the numbers and the peer review.
V2 line 20 argues for "embedding moral and ethical frameworks" at the substrate level, not layering them on afterwards as filters. This is the embedded-alignment thesis. The name Eden Protocol is in this bundle as well. The message body reads "The theoretical framework for the Eden Protocol, designed to embed moral intelligence into AI from inception", and the attached book draft v3.2 carries "Chapter 20: The Eden Protocol: Intelligence Born in the Garden". Anyone who wants to check that can search the.eml text; the word occurs in the message body and in that attachment, and not in the Version 2 essay.
V2 lines 6 to 9 state U equals I times R. Papers I and II formalise this as U equals I times R to the alpha, and the published book presents the alpha of approximately 2 case as U equals I times R squared. Priority on capability-growth scaling is claimed carefully: the equation is cited with its artefact, and the fixed-exponent framing as the operative safety criterion is superseded in Paper X by the beta greater than k co-scaling condition. The equation itself and the ARC Bound remain live hypotheses; only the earlier unblinded single-model fit of alpha approximately 2.24 was retracted, corrected to approximately 0.49 under blinding.
V2 line 189 calls for "governments, scientists, and religious leaders" to collaborate. This is the interfaith governance call that the register maps to the Pope Leo XIV encyclical in May 2026 and to Christopher Olah co-presenting at the Vatican Synod Hall on the same day. Priority on either external event belongs to those institutions.
Babylon and Eden do not appear. Alignment faking as a mechanism is not described. Reinforcement learning specifics are not described. Any claim that the manuscript predicted the details of the Anthropic experiment would be false, and the truth gate blocks it. What is in the manuscript is the direction of travel, in language, with a timestamp that no later editing can move.
Open the.eml. Verify the Message-ID. Recompute the SHA-256. Extract the text of the attachments and search for the lines quoted here. The register either survives that check or it comes off the register. The earlier unblinded single-model fit of alpha approximately 2.24 did not survive its own check; it was retracted and corrected to approximately 0.49 under blinding, and its retraction sits at the top of the falsification dashboard as a reminder of what the honesty in this programme is for.
Related notes: the ten days that reset three fields for the fortnight of 8 to 18 December 2024; April 2025: the manuscript that expanded the Eden Protocol for the successor artefact; Claim 1 explained for the embedded-alignment thesis; Claim 17 explained for the HRIH cosmology (speculative). New readers arriving cold: /start-here.html is a one-page orientation to Michael Darius Eastwood, the ARC/Eden research programme and the priority record.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.