What happened
BeInCrypto reported Friday that a developer running OpenAI's GPT-6 Astra pointed the model at a Nazi-era Enigma ciphertext and produced a readable plaintext in roughly 10 hours of compute. The archive attached to the write-up logs the prompts, intermediate outputs, and the final decrypt, which turns the claim from a tweet-thread flex into something other researchers can pick apart.
Enigma, for context, is a mechanical rotor cipher used by German forces in the 1940s. Its keyspace looks large on paper but collapses fast under known-plaintext attacks and modern compute, which is why it has been a favorite teaching target for cryptanalysis classes for decades. What is new here is the operator, not the target.
A general-purpose large language model, not a bespoke cryptanalysis pipeline, sat in the driver's seat and coordinated the attack end to end.
Why it matters
The headline reads like a cryptography story. The subtext is about capability trajectory. Every time a frontier model chews through a task that used to require specialized tooling, the market for post-quantum and post-AI cryptography assumptions gets a fresh look.
Bitcoin sits inside that conversation whether it wants to or not. The chain leans on SHA-256 for proof-of-work and mining, and on ECDSA over secp256k1 for spending signatures. Neither is comparable to Enigma in structural strength.