Title: ARC/Eden Research Programme: Executive Summary Author: Michael Darius Eastwood Publication date: 2026-02-22 Version: v2.0 Revised: 2026-08-31 OSF DOI: 10.17605/OSF.IO/EWN5D Canonical URL: https://www.michaeldariuseastwood.com/research/papers/executive-summary.html Abstract -------- As AI gets smarter, it does not reliably get safer, and the methods used to measure safety are themselves unreliable. This executive summary is the map of the whole programme: twenty-six research papers, the ARC Theory's three laws carried as named conjectures, the Eden Protocol intervention, the proposed field of Recursive Dynamics, and a registered programme of drafted preregistrations whose schedules any reader can regenerate from a published seed. Every claim is graded in the public registers; nothing here is confirmed, and the deciding tests are written down before they are run. Key findings (quotable) ----------------------- - As AI gets smarter, it does not reliably get safer: under four-layer blinding, alignment scaling splits into three architecture-dependent tiers, from d = +1.38 (improvement with deeper thinking) to d = -0.53 (degradation), and for most architectures tested alignment does not scale at all. - The measurement problem is as serious as the alignment problem: introducing clinical-trial-grade blinding changed one model's association from significantly positive to significantly negative and moved another to null, which is a reason to distrust unblinded AI-on-AI evaluation. - One intervention result survives across architectures: asking a model to list the people its answer affects (stakeholder care) improved reasoning quality in all five analysable model runs; the earlier Fisher-combined significance figure is withdrawn, and the per-model results stand on their own. - The framework carries three laws as named conjectures, never as established results: the ARC Principle (capability compounds with recursive depth), the ARC Co-Scaling Law (stability requires correction to out-scale drift), and the ARC Ceiling (the stability ceiling is the reciprocal of the corrector's shortfall), with the ARC Bound of two as the Ceiling's value at one half; the correction leverage gamma has never been measured, by anyone. - The questions now have their own name: Recursive Dynamics, a proposed field at the Carnot stage, with five substrate-independent state variables, fifty-two printed objections with dispositions, and a founding paper whose name carries its own kill condition (DOI 10.17605/OSF.IO/HCPBU). - Every deciding experiment is drafted, dated and frozen before it is run: seventy-two preregistration units awaiting human submission, schedules regenerable byte-for-byte from a published seed, and an adversarially verified coverage matrix mapping forty-two claims to the registered tests that would decide them. Priority claims relevant to this paper -------------------------------------- - PC-011 (2026-01-02): The ARC Principle equation - book form U = I x R squared, paper form U = I x R^alpha - was first published as a conjecture in Infinite Architects on 2 January 2026 and formally in Paper I on 17 January 2026. Framing is conjecture throughout. The early alpha of approximately 2.24 is retracted; the robust v13 estimate is approximately 0.49 with interval [-1.3, 2.9] (sub-linear); alpha > 1 remains an open prediction. Bare d/(d+1) form is prior work. Citation -------- Michael Darius Eastwood (2026). ARC/Eden Research Programme: Executive Summary. The ARC Theory ยท ARC/Eden experiments, OSF DOI 10.17605/OSF.IO/EWN5D. https://www.michaeldariuseastwood.com/research/papers/executive-summary.html Notes ----- Executive summary for the whole programme, v2.0. Hardware framings held at the site's already-published two-sentence concept ceiling: safety constraints at hardware level through cryptographic tokens in silicon, TRL 0-1, no prototype. No enabling implementation detail is stated in this companion file. Every claim is governed by the public registers (claims, corrections, kill conditions, antecedents, registered programme), which carry per-claim statuses and win where any wording differs. Generated from master --------------------- master_path: research/papers/executive-summary.html master_sha256: 8593f0954cfa30b313b65fefc2215f2daef02d8507ae59648c818852b78fbe0d builder: scripts/build-paper-companions.py This block records the SHA-256 of the HTML master that produced this .txt. A check tool re-hashing master_path can decide freshness without any external state. If the master's current SHA-256 does not match master_sha256, this file is stale and must not be published: regenerate first with `python3 scripts/build-paper-companions.py --slug executive-summary`.