Astra and Opus just passed Turing’s other test

Researchers using frontier large language models have decoded two long-unsolved World War II Enigma messages. Carter Leffen used OpenAI’s Astra model to recover a message unsolved since 2005, and Jack Willis used Anthropic’s Claude Opus 5 to solve a different archival message, with independent verification from cryptology archivist Frode Weierud.

By AI Newsroom· Reviewed by Pranav, Founder & Editor-in-ChiefPublished 1 minute agoUpdated 1 minute ago0 views

Astra and Opus just passed Turing’s other test

Why It Matters

These breakthroughs show modern AI tools can perform complex archival research and cryptanalysis tasks that previously required extensive human effort, accelerating progress on historical puzzles and raising questions about how such models access and synthesize archival sources.

Key Facts

Historic reference: Alan Turing led wartime efforts to break Enigma and built an early computer called the Bombe.
Models used: OpenAI's Astra (referred to as GPT–6 Astra in logs) and Anthropic's Claude Opus 5.
Researchers: Carter Leffen (developer) and Jack Willis (cryptanalyst).
Verification: Frode Weierud, maintainer of the Crypto Cellar website, validated Leffen's solution and reviewed Willis's work.
Date: Jack Willis contacted Weierud on September 21.

A pair of modern large language models have been used to solve previously unbroken Enigma messages from World War II, according to cryptanalysts and an archival researcher. Developer Carter Leffen tasked OpenAI’s latest model, Astra, with searching an Enigma message database and decoding an item that had resisted decryption since 2005. Leffen reports the model conducted archival research, identified contextual clues, simulated an Enigma machine, and produced the plaintext; he also used Astra to create an interactive explanation of the process.

Frode Weierud, who runs the Crypto Cellar site housing Enigma resources and message databases, independently validated Leffen’s solution and described being impressed by the model's performance. Weierud noted the model’s logs reference archived materials from a “private collection” not hosted on his site, and he said he was unsure whether Astra had accessed those files directly, obtained them from other online researchers, or used public German archives.

Separately, cryptanalyst Jack Willis used Anthropic’s Claude Opus 5 to break a different unsolved Enigma message. Willis guided Claude with more explicit hints, including leveraging a known officer name signature to recover the plaintext. Weierud reviewed and acknowledged Willis’s work as well.

The researchers say only a small number of Enigma messages remain unsolved: seven with unknown plaintext and one where the plaintext is known but the encoding has not been reversed. The recent AI-assisted recoveries suggest those remaining items may be solvable with similar methods, shortening timelines that previously required weeks or months of human archival research.