Artificial intelligence has just reached a symbolic milestone in historical cryptanalysis. Thanks to the computing power of the GPT-6 Astra model, a German military telegram sent on November 27, 1918, which had remained indecipherable for over a century, has finally revealed its contents. Using the ADFGVX encoding system, considered the most robust field cipher of World War I, this 170-character message described Allied ship movements in the port of Sevastopol. The accuracy of this decryption was empirically confirmed through cross-referencing with digitized Royal Navy logbooks, which attest to the exact presence of the British vessels on the specified date.
Cracking this code relied on discovering the transposition key TRUPPENVERSCHIEBUNG, a 19-letter German word. The long-standing failure of cryptanalysts was primarily due to a time discrepancy: archival records listed the official use of this keyword starting on December 9, 1918, twelve days after the message was sent. By testing statistical hypotheses based on letter frequency and language structure, the algorithm overcame this chronological anomaly. Although about a dozen similar messages still resist researchers, this success illustrates the effectiveness of large language models in identifying complex textual patterns hidden beneath manual encryption methods.
This breakthrough reignites the debate over the vulnerability of current security systems, particularly those protecting decentralized networks like Bitcoin. However, the method used to break a historical code does not apply to modern standards. The architecture of the original cryptocurrency relies on SHA-256 hashing and the secp256k1 elliptic curve, which depend on 256-bit private keys generated in a completely random manner. With a search space of approximately 1.16 × 10⁷⁷ combinations devoid of any linguistic or statistical footprint, these algorithms remain completely immune to the pattern-analysis capabilities of deep learning models.
The real challenges for the digital financial ecosystem do not stem from advancements in conversational agents, but from future developments in quantum computing. While recent tests have demonstrated the feasibility of cracking very short 15-bit keys on experimental quantum architectures, the real threat to ECDSA signatures lies in the theoretical application of Shor's algorithm on a large scale. It is this technological timeline, tied to quantum resistance, that commands the attention of developers and financial institutions today, far more than the evolution of natural language processing models.