OpenAI’s Astra and Anthropic’s Opus break two unsolved Enigma messages
Two cryptanalysts used OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5 to decode two long-unsolved Enigma-encoded messages from World War II. Carter Leffen prompted Astra to find and break a message that had stumped researchers since 2005. The model located archival clues, simulated the Enigma machine, and recovered the plaintext in days—work that would take humans weeks or months, according…
Key points
- OpenAI’s GPT-6 Astra decoded a 2005-unsolved Enigma message in days, using archival research and simulation
- Anthropic’s Claude Opus 5 broke another message by identifying an officer’s known signature pattern
- Only seven Enigma messages remain unbroken, per Crypto Cellar’s Frode Weirerud
Jack Willis used Opus 5 to break a second unsolved message by leveraging a known officer’s signature. Weirerud confirmed only seven Enigma messages remain unbroken, plus one with known plaintext but unresolved code. The demonstrations highlight how advanced LLMs can replicate complex historical research tasks with speed and precision.
Model page: GPT-6 (Astra) →
The story so far
2 episodes →- OpenAI’s Astra and Anthropic’s Opus break two unsolved Enigma messagesthis story
Astra and Opus just passed Turing’s other test
TechCrunch AI · 25 September 2026
Loading the full article…
This text was published by TechCrunch AI and written by Tim Fernholz. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- OpenAI fires contractors for using AI to train its models · 2 src
- OpenAI’s research chief discusses AI safety and self-improvement goals · 1 src
- AWS boosts MoE RL training throughput by 40% with EFA and DeepEP on EKS · 1 src
- AWS shows how to train Qwen3-VL-8B with SkyRL on SageMaker HyperPod · 1 src
- Researchers unveil UpTCR model for TCR-antigen binding prediction · 2 src
Comments
via GitHub Discussions