GPT-6 Astra cracks WWI German radio cipher
| Source: HN | Original article
GPT-6 Astra has decoded a World War I German radio cipher, marking a breakthrough in applying AI to historical cryptanalysis.
OpenAI’s flagship model GPT‑6 Astra has cracked a World War I German naval radio cipher that has eluded historians for decades. Two independent builders reported that Astra deciphered a message transmitted on 27 November 1918 using the ADFGVX field cipher, a complex system employed by the German navy in the war’s final weeks. The plaintext generated by the model matches entries in British naval logs – notably the HMS Canterbury’s records – providing the first credible reconstruction of the transmission’s content.
The breakthrough was disclosed on a public list of roughly fifty historically unsolved ciphers maintained at scienceblogs.de. One of the builders, publishing under the name “prinz,” described the process as a “plausible plaintext” that aligns with contemporary British records. A separate report on the explainx.ai blog noted that the same model had earlier solved a 1941 German Army Enigma message, a result that received direct sign‑off from an Enigma historian. Together, these cases illustrate Astra’s emerging strength in cryptanalysis, a field traditionally dominated by human experts and specialized algorithms.
Why it matters is twofold. First, the result showcases the practical power of large language models to tackle pattern‑heavy problems beyond natural language, potentially accelerating research in historical linguistics, archaeology and intelligence studies. Second, it raises policy questions about the accessibility of advanced AI tools that can decode legacy encryption schemes, prompting discussions on responsible disclosure and the safeguarding of sensitive historical data.
What to watch next: OpenAI is likely to field further demonstrations as developers experiment with Astra on the remaining unsolved messages from the scienceblogs.de list. Academic teams may seek formal collaborations to validate AI‑generated decryptions, while regulators and security experts will monitor how such capabilities are shared and controlled. As we reported on Astra’s phased rollout in mid‑September, the model’s rapid move from legal‑research applications to historical codebreaking underscores its versatile, and sometimes unexpected, impact.
Sources
Back to AIPULSEN