Paper III · The ARC Theory · Law II, the ARC Co-Scaling Law · The Alignment Scaling Problem: Why External AI Safety Approaches Cannot Scale With Recursive Capability

This paper reports alignment scaling exponents statistically indistinguishable from zero on the systems measured, which would mean safety degrading relative to capability as recursive depth increases; the registration records that reading as contestable, since a materially positive but below-threshold exponent would refute it as worded. If AI capability scales super-linearly through recursive self-correction (demonstrated earlier and independently in 95.6% of tested configurations), but alignment constraints such as RLHF, constitutional rules, and output filters operate externally to the recursive reasoning process, the safety ratio degrades as depth grows. Canonical version and all formats: https://www.michaeldariuseastwood.com/research/papers/paper-iii-alignment-scaling-problem.html DOI: 10.17605/OSF.IO/HQCGF Author: Michael Darius Eastwood, ORCID 0009-0004-3222-7442, https://michaeldariuseastwood.com Licence: CC BY 4.0 Michael Darius Eastwood conceived and directs this research programme and is the author of this work. Across the programme, he has used more than six AI systems in parallel, under his own instructions, to stress-test his arguments, identify possible errors, and assist in preparing draft text from his own outlines. He determines what is adopted, revised or rejected and takes responsibility for the published content. These systems are tools, not authors.