The ARC Theory
The claim, in one sentence: a system that improves itself can only stay correctable up to a limit set by how fast its own correction improves. The sharpest test of any single link is written, dated and not yet run; the claim itself is proposed and unproven, and no single measurement decides a law.
If you are choosing where to start: the dated record says where each prediction stands, the evidence is graded row by row, and the book is free to read in full. Research proposals, author-run results and independent validation are different things, and I keep them apart.
No institution employs me and no grant pays for this. The dates are on the dated record, checkable tonight on your phone, so judge the record, not the person keeping it. A single-investigator, evidence-gated programme testing one idea: that capability in recursive systems scales with recursive depth, and that alignment must therefore be embedded in the loop rather than imposed from outside. Every claim is dated, hashed and publicly anchored. Results publish with kill conditions fixed in advance, corrections are logged in public, and the whole record invites independent replication.
The chain, paper by paper: recursive processing → capability scaling → the alignment scaling problem → the limits of external checking → embedded correction → correction that must scale with what it corrects. Each link is a separate paper with its own kill condition; judge the chain, not one link.
How it fails, register by register
The theory’s scientific content is not all-or-nothing: it partitions into independently falsifiable hypothesis registers, and the failure of one is not evidence against another unless the stated dependency holds. HRIH, the cosmological wing, is a separate hypothesis outside the empirical core: nothing below requires it, and no outcome below is evidence for or against it.
| Register | What kills it | If it falls |
|---|---|---|
| ARC-1 · Capability scaling (Law I) | the depth-scaling re-measurements failing to recover the stated relationship | ARC-1 falls; the persistence registers stand |
| ARC-2 · Correction co-scaling (Law II) | sustained self-improvement with falling correction and no drift | ARC-2 falls and takes ARC-3’s derivation with it |
| ARC-3 · The ceiling (Law III) | a cross-against-same corrector difference wholly inside the equivalence region fixed in the draft registration; or a same-class correction exponent wholly above one half, felling the substrate cap | ARC-3 falls; the co-scaling law stands on its own result |
| ARC-4 · Value persistence | embedded loops faring no better than removed enforcement in the persistence trials | ARC-4 falls and takes ARC-5 with it |
| ARC-5 · The genesis strategy | formation failing to outperform enforcement after release | ARC-5 falls; ARC-4 stands as measured |
| The priority | an earlier document holding the claim’s first four components | the priority falls; no law falls with it |
Stated now, so nothing can be moved later. None of the decisive trials has been run; the theory has never yet been at risk, and it says so.
Every kill condition and its live status is in the falsification register. This section is drawn from the two-page form, the supporting paper of The ARC Theory: Statement Paper.
No paper in this programme is peer reviewed. All are dated preprints on OSF carrying DOIs and hashed priority records, and the site grades every claim as stated, dated and unproven wherever it appears.
The apparatus, drawn: eight verification surfaces, from the sealed record to public self-corrections
What this programme is
27 dated public documents and 1 supporting paper: the statement paper, the empirical series and the supporting frameworks, spanning AI reasoning, quantum error correction, classical physics and the mathematics of scaling. They sit on one OSF project with a DOI of its own, and the individual papers carry their own DOIs, so a citation can point at the whole record or at a single document. The papers are the argument; the evidence infrastructure beneath them is the product. Nothing is asserted without an artefact you can check.
What it believes could help the world
The directional thesis makes three claims: recursive processing systematically outperforms parallel, alignment cannot be imposed from outside, and capability and alignment scale independently. If those hold, AI safety needs to shift toward embedded-alignment architectures. This programme exists to make that case with evidence, not persuasion.
The premise is West's. The question is Chalmers' and Hutter's. What is claimed here is the replacement limit, its number, and its dependence on architecture.
The empirical alignment series · Papers I to XIII
Start here: the Executive Summary is the map of the whole programme in one document, and the Master Table of Contents lists every paper, format and DOI. The series below is the evidence itself.
The statement paper: the whole claim in one document, law before value, with the dated record, the concessions and the refutation conditions.
The founding paper of the proposed field: five state variables, three laws as named conjectures, the ARC Ceiling derived in full, twelve hosts tried…
This paper formalises and preliminarily tests the ARC Principle (Artificial Recursive Creation), first proposed in Infinite Architects (Eastwood…
This paper presents experimental validation of the ARC Principle across multiple frontier AI models (Claude, DeepSeek, Gemini, Grok, Groq Qwen, GPT)…
This paper reports alignment scaling exponents statistically indistinguishable from zero on the systems measured, which would mean safety degrading…
This paper presents evidence that frontier language models fall into distinct alignment response classes when inference-time reasoning depth is…
This paper analyses the relationship between inference-time reasoning depth and ethical reasoning quality using the ARC Alignment Scaling…
This paper presents ARC-Align, a blind benchmark for evaluating AI alignment quality as a function of inference-time reasoning depth. Current…
One apparent positive result went flat, another turned negative, a third held flat, when blinding and other protocol components changed together.
This paper presents empirical evidence from a six-model Eden Protocol intervention suite (Claude Opus, DeepSeek, Gemini Flash, Grok Fast, Groq Qwen…
Simulation evidence that embedding safety inside the optimisation objective of a self-modifying system (capability multiplied by safety, the honey…
Cauchy's four functional equations -- additive, multiplicative, exponential, and logarithmic -- are the only continuous solutions to their respective…
The assumption that AI safety imposes a capability tax has shaped alignment research for a decade. It has also created the single most dangerous…
The synthesis paper for the ARC/Eden research programme. The synthesis paper for the programme; integrates the papers that precede it into a single…
The coupled co-scaling correction for keeping AI systems inside a safe corrective envelope.
A dated, graded convergence record across AI, quantum computing, hardware, doctrine and mathematics; every row carries its primary source and grade…
Within the ARC Theory: the external-validity protocol of the blinding law on public benchmarks.
A derivation, not a measurement: the notation collision resolved and the measurable threshold joining recursive capability growth to correction.
Working paper v1.1: the ARC programme's forward predictions assembled as one dated, verifiable chain, with the ensemble doctrine, the graded book…
Supporting papers
Shorter documents that support a paper of the programme and cite it for their claims. Each carries its own DOI and states which paper it supports.
The whole claim in one sitting: question, five claims, the three ARC Laws graded, dates, kill conditions, retraction and invitation.
Beside the series
Distinct axes alongside the empirical series: the cognitive profile behind the work, and the original speculative hypothesis, preserved and dated.
Paper C: a falsifiable descriptive framework for radical cognitive asymmetry, the Capability-Adjustment Fallacy, a documented live case across…
Labelled speculation, held apart from the argument and severable from it: where does recursive self-improvement end? A testable creation theory: if…
Supporting frameworks
In the past eighteen months, a versioned set of related external research developments have discovered that recursive or recurrent processing produces capability…
A mouse's heart beats 600 times per minute; an elephant's beats 28; a blue whale's beats 6. This paper traces the origin of scaling laws across…
Every engineering project begins with requirements. Every set of requirements begins with purpose. And every purpose, traced back far enough, arrives…
Withdrawn from circulation on 14 July 2026 pending disclosure review. A withdrawal notice is retained at the canonical address.
As AI gets smarter, it does not reliably get safer, and the methods used to measure safety are themselves unreliable. This executive summary provides…
This document provides a master table of contents, reading guide, timeline, key results summary, and glossary for the complete ARC Principle paper…
Evidence infrastructure
The verification layer. Every claim links to its source; every source is hashed.