What did the adversarial review of the ARC Theory find?
Michael Darius Eastwood's adversarial precedence review set ten frontier AI configurations from six vendors to attack his novelty claims. It found the components in earlier work but no single earlier work joining them, and named the operationalisation, turning the theory into measurements that can fail in public, as the strongest surviving claim. He commissioned it; it is not independent review.
Text reviewed 2026-10-31. Any earlier audio edition may differ; the dated text and linked records govern.
Its concessions are published on the synthesis and related-work pages, and the book had already disclaimed the components in print: "I make no claim to have invented the components." and "What I offer is the synthesis" (Infinite Architects, Author's Note). The two most load-bearing philosophical joins were found substantially anticipated and are conceded by name, including the prior argument on the difficulty of control (Yampolskiy, 2020).
The review narrowed the priority claim to the operationalisation and the named relations, and it does not decide whether any of them is true: a well-specified test can still find a theory wrong. Even on its strongest row the word first is not asserted, because the formal systematic literature search that would license it has not been run. Three structural claims made after the review are recorded as unsearched, not as firsts.
It was told to be hostile, it used AI systems, and the programme commissioned it, so it is commissioned adversarial review, not independent external review. Its use is the specific criticism it produced, each item checked at its primary source and carried into the papers or the corrections record.