TechCrunch · AISep 26, 2026, 01:24 UTC+8

OpenAI Astra and Anthropic Opus just cracked unsolved WWII Enigma messages

Astra and Opus just passed Turing’s other test

Two cryptanalysts used OpenAI's Astra and Anthropic's Opus to decode two Enigma messages that had remained unbroken since WWII. Developer Carter Leffen had Astra search archives, find context clues, build an Enigma simulator, and recover the plaintext. The post doesn't spell out Opus's exact role, nor the time taken or accuracy rate.

Why it matters: The story has strong narrative pull and a concrete knowledge hook in Astra's autonomous simulator-building. But Opus's role and key metrics are missing, and historical codebreaking is far from daily AI workflows, capping the score at the featured threshold.

Read the original ↗Export Markdown

Take

I clicked because it puts AI into Turing's old job: cryptanalysis. Developer Carter Leffen had Astra search archives, find context clues, build an Enigma simulator, and recover the plaintext. It reads like a clean demo of chained tool use—search, reason, code, verify—all wired together.

I'd discount it a bit, though. The post barely mentions Opus's contribution beyond "also participated." Was it independent decoding, cross-validation, or just grunt work? And the two numbers I actually want—time taken and accuracy—are missing. Did it crack the message in seconds or three days? 100% recovery or half-right guesses? Without those, it's hard to tell if this is a real capability signal or a well-tuned demo.

The direction is interesting, though. Enigma messages have known encryption rules and tons of historical archives, which makes them a decent testbed for long-chain tool use. If someone publishes the full process and benchmarks, it'll be more convincing than the headline alone.

Source: TechCrunch · AI · techcrunch.com
Published: Sep 26, 2026, 01:24 UTC+8 · 4 days ago