Michael Darius Eastwood · Independent AI alignment researcher
Published
Fifteen days after the ebook went live, the author self-emailed a curated evidence bundle. This is the entry the register catalogues as email 4 of 8 and it is worth explaining because it is often misread as a convergence. It is not. It is a record.
Four things, all of them checkable by anyone. The Message-ID and SHA-256 of the 8 December 2024 manuscript. The Message-ID and SHA-256 of the 30 April 2025 manuscript. The ISBN of the published book. And the two convergence entries that had already fired in December 2024, with their primary source citations: arXiv:2412.14093 for Anthropic's alignment-faking paper and the Nature paper for Google Quantum AI's Willow chip.
By 22 January 2026, the material had been in front of a mainstream audience for two weeks. A single-email bundle of hashes, IDs and primary sources gives any curious journalist a starting point for verification that does not depend on the author's account of anything. The bundle contains no new claims. It contains no new results. It restates in one place what had already been made public across the previous thirteen months.
The register's rule for engagement rows is explicit. Any MDE-initiated contact is listed separately from convergence rows and is not counted toward the headline number of 19 independently sourced convergences. This bundle is engagement, and the register logs it as such. The convergence count is 19 because those 19 events happened externally, by other institutions, from independent starting points. The engagement rows exist so a reader can see the outreach separately and judge it separately.
No reply has been received to date. The methodology on silence is explicit: silence in reply is silence. It is not evidence that the material is correct and it is not evidence that it is wrong. The programme's rule against citing silence as endorsement is enforced by the truth gate. This distinction sounds pedantic and is not; a research programme that treats non-response as agreement is a research programme that has stopped listening.
The .eml is downloadable. The SHA-256 begins af863bfa. Every hash and primary source cited in the bundle is independently checkable: the arXiv paper, the Nature paper, the Gmail Message-IDs for the two earlier manuscripts, the ISBN through any library catalogue. The bundle survives its own audit or it does not, and the programme prefers the answer to be public either way.
Records like this exist for the same reason preregistrations do. They fix a claim in place at a moment in time, so that later versions of the story cannot quietly move that moment. The 22 January 2026 bundle joins the falsification dashboard and the retraction protocol as one of the programme's disciplinary artifacts. It is not exciting. That is largely the point.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.