OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005

	OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005(cryptocellar.org)
	725 points by sohkamyung 21 days ago \| 437 comments
	tl;dr: OpenAI's GPT-6 Astra autonomously broke a German Army Enigma message (MVUEH, sent 10 July 1941) that had resisted cryptanalysis since 2005. Given only a general prompt to attempt unbroken messages from the Crypto Cellar Research site, the model identified MVUEH as promising, hypothesized a link to a previously broken related message, wrote its own Enigma simulator and Bombe in Python/C++, and used "ROSENOW ROSENOW" as a crib to recover the key. Logs also show it independently located relevant Bundesarchiv file references, performing weeks of cryptanalytic and archival work in two days.
	HN Discussion: •Provides the actual decrypted message content as helpful context ~Other LLMs like Gemini can replicate the decryption easily, diminishing the achievement's uniqueness ↓Claim of autonomous solving is overstated since much work was offloaded to generated software and human collaboration ↓Title is misleading; it was a human-AI collaboration, not an AI solo achievement ~LLMs can solve problems similar to previously-solved ones but likely lack original thinking for truly novel problems