GPT-6 Astra traces Newton alchemical source and 1941 Enigma NF messages

Frontier models solved two constrained historical problems this week. GPT-6 Astra matched a previously unattributed Newton Latin passage to a French alchemical original. The same model parsed July 10 1941 Enigma traffic by cross-referencing newly digitized SS-Totenkopf Nachschubführer cleartext logs released in 2026. Both outputs were produced in single autonomous runs against fully digitized corpora.

Success required four conditions stated in the source: expert-curated open problems, machine-readable primary sources, verifiable ground truth, and code-generation capability. The Enigma break hinged on Astra surfacing the NF numbering note rather than cryptanalysis itself. The Newton identification hinged on multilingual alignment across 17th-century printed editions. Neither result was reachable by last-generation models.

These cases demonstrate operational limits. Only problems with clear disproof criteria and complete digitization qualify. Cryptography and translation tracing meet the threshold; diffuse interpretive debates do not. AI labs and archives that release structured corpora will capture the next increments.

Funding agencies should issue targeted calls for historian-model pairs on cipher collections and alchemical corpora by Q2 2027. Deliverables must include machine-readable proof logs and external historian verification within 90 days.