Michael Darius Eastwood · Independent AI alignment researcher
Published
This note collects the arc of the ARC/Eden research programme in one piece. It is not a promotion; it is a chronology with the dates the register treats as load-bearing. A reader who has followed row by row can skip it. A reader arriving cold can use it as a map. New to the programme? Start at /start-here.html for the five-minute orientation, then come back here for the timeline.
8 December: a Google-server-timestamped self-emailed manuscript arguing that AI systems cannot be truly controlled and that correction must be embedded at the substrate. Word count 189,355; SHA-256 f0d1f38f. 9 December: Willow, the first below-threshold quantum error correction (Google Quantum AI, Nature). 18 December: alignment-faking reasoning at 78% in the RL-training condition (12% baseline), Greenblatt and colleagues, arXiv:2412.14093. Three fields, three institutions, ten days.
30 April 2025: a 562-page manuscript, self-emailed. The Babylon and Eden framing first appears, and the embedded-alignment thesis is now named the Eden Protocol. Caretaker Doping, Meltdown Triggers, and Meltdown Alignment enter the register. SHA-256 anchor prefix 09f5b5e1. Around the same period, FlexHEG appears from Petrie and colleagues at ARIA, a tamper-proof guarantee-processor concept for AI chips, arXiv:2506.15093. Odeh files hardware-level ethics-enforcement patents in July, published early 2026.
September 2025: Petrie publishes embedded off-switches for AI compute, deadman-switch security blocks per accelerator at less than one per cent die overhead, arXiv:2509.07637. Concurrent with the Meltdown Triggers framing in the April manuscript. November 2025: Sharma and Chopra publish the sequential-parallel result, sequential greater than parallel in 95.6% of configurations, arXiv:2511.02309. Priority on that quantified finding is credited to them; the directional thesis appears in the December 2024 manuscript.
The book is published, paperback 2 January (ISBN 978-1806056200; ISBN-10 1806056208), ebook 6 January (ISBN 979-8302196163; Kindle ASIN B0GDQZ7P21). It introduces the Chokepoint, the HARI Treaty, the Three Ethical Loops, and the U equals I times R squared presentation of the ARC Principle. 7 January: an outreach email to BBC's Faisal Islam. 17 January: OSF project created and Paper I released, DOI 10.17605/OSF.IO/6C5XB. 22 January: Paper II released and a self-emailed BBC evidence package.
February: NYU publishes classical time crystals via nonreciprocal forces, PRL 136, 057201; Zhang and colleagues publish the first acoustic topological space-time crystals, Newton 2, 100334. 24 March: outreach emails to Professors Demetrius (Harvard) and West (SFI); no reply received to date. 14 April: PNAS Nexus publishes Hernández-Espinosa and colleagues on neurodivergent influenceability as a contingent solution to alignment, DOI 10.1093/pnasnexus/pgag076. 17 April: Recursive Superintelligence closes 500 million US dollars at 4 billion US dollars, GV plus Nvidia. 25 May: Pope Leo XIV's first encyclical Magnifica Humanitas, Babel versus Jerusalem framing, with Christopher Olah of Anthropic co-presenting at the Vatican Synod Hall.
3 June: EU Tech Sovereignty Package, Chips Act 2.0 and CADA, with Virkkunen naming a kill-switch concern as the framing motive. 12 June: US orders Fable 5 and Mythos 5 offline, the first export-control action against a deployed commercial AI model. 19 June: Macron at G7 Évian, strictly-nationalist critique and the trusted-partners framework. 24 June: OpenAI plus Broadcom's Jalapeño inference chip announcement. 26 June: Gumbau Mezquita's undecidability proof and the Trakhtenbrot's Wall trilemma, arXiv:2606.28639.
A measurement programme with a preregistration path and a 624,000 pound ask. Nineteen convergences do not settle the science. They make the shape of the argument visible to a reader who does not have to trust anyone in it. That is what the arc was for.
New readers: /start-here.html for the five-minute orientation to Michael Darius Eastwood and the research programme, then Paper I in plain English. Followers of the priority record: /priority-evidence.html for the verbatim quotes and hashes, and /research/priority/ for the machine-readable chain. Anyone verifying: recompute the SHA-256 prefixes f0d1f38f and 09f5b5e1 from the redacted .eml files under evidence/emails/, then compare the extracted text to the quotes on the priority evidence page. The register is either survivable or it comes off the register; the choice of which is not the author's to make.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.