Tom Di Mino, a self-taught AI engineer from New York’s Hudson Valley, says he has done something linguistics scholars have failed to do for over a century: read Linear A, the Minoan script used in Crete between roughly 1800 and 1450 BC. His methodology is documented, his outputs are concrete, and his claims are currently under review by academics at Rutgers and Cambridge. The central question the field will need to answer is not whether Di Mino is interesting. It is whether what he found is a genuine decipherment or a sophisticated pattern match that happens to fit.
That distinction matters enormously. Undeciphered scripts attract a long tail of claimed solutions. Gordon’s 1957 Semitic argument for Linear A was published in the journal Antiquity and never gained acceptance. The graveyard of confident Minoan decipherments is large enough that professional linguists treat new claims with calibrated skepticism by default.
Di Mino’s approach is different from prior attempts in one important way: he used software to do the work at scale. He built a suite of Python scripts with Claude Code, Anthropic’s agentic coding tool, to query and cross-reference the GORILA and SigLA databases, the two main digitized archives of Linear A inscriptions. That let him run hypothesis tests across the full corpus rather than working inscription by inscription. The computing infrastructure is not the claim, but it is what made the claimed insight operational.
The specific breakthrough Di Mino describes came on May 22, 2026, while he was analyzing prayer inscriptions from five Minoan sanctuary sites. Each inscription followed the same formula, with every word in the line identified by its overlap with Linear B, the Mycenaean script that shares 60 symbols with Linear A and was deciphered in 1952 by Michael Ventris. The exception was the first word: a verb root that appeared across all five sites in regional variants but could not be read.
Di Mino focused on a Linear A-only sign, labeled in the corpus as *301, and proposed that it carried the phonetic value “na.” With that value in place, the verb root became “nawaya,” meaning “to dwell.” He argues that N-W-Y as a three-consonant Semitic root for dwelling appears in Hebrew, Akkadian, and related languages, and that placing this reading into the prayer formula produced text structurally similar to known early Hebrew liturgical forms, addressed to a goddess rather than a deity of the Hebrew tradition.
From that pivot, Di Mino has produced a proposed phonetic reading for 40 signs (including 13 with no previously accepted value), a lexicon of 408 translated Linear A terms, and a nine-page manuscript draft titled Ya Diktu: Grammar of the Minoan Peak Sanctuary Libation Formula, which he may submit for peer review.
The scope of what he has produced is notable. What is not yet in evidence is independent verification. The account of Di Mino’s work was published on aiclambake.com by a writer who knows him socially, an unusual provenance for a claim of this magnitude, and the author flags this themselves. The Rutgers and Cambridge review is in progress, not complete. The manuscript is nine pages and pre-submission. The 408-word lexicon and the 40 sign readings have not been replicated by a second researcher working from the same premises.
None of that means Di Mino is wrong. It means the evidence currently available is one researcher’s documented hypothesis, not a confirmed decipherment. The history of Linear A scholarship gives particular reason for caution around Semitic hypotheses, not because the theory is impossible but because earlier Semitic arguments also appeared internally consistent and did not survive external scrutiny.
What Di Mino has demonstrably done is use computational infrastructure to systematize a hypothesis-testing workflow that would have taken years manually. If the Semitic reading holds under peer review, the tooling will have mattered as much as the linguistic insight. If it does not hold, the same tooling will have produced an unusually well-documented wrong answer, which is still more useful than an undocumented one.
Researchers and journalists covering this claim should wait for the Rutgers and Cambridge verdict before treating the decipherment as established. The interesting story right now is not that Linear A has been read. The interesting story is that a solo engineer with access to digitized corpora and an agentic coding tool got further in five months than most credentialed linguists get in a career.
Reported by aiclambake.com, published June 16, 2026. The author of the original post notes a social relationship with Di Mino and that the expert review is ongoing.
