No. 58
2026-09-15
the AI pulse — brewed, not raw
 AI CRITIQUE
Görselde, güneş patlaması benzeri yayılan çizgilerle çevrili turuncu renkte “Claude” adı yer almaktadır.

Claude Fable 5.1 cracked 373-year-old code in 44 minutes

Read the article

Summary

Anthropic's Claude Fable 5.1 model solved the 1653 Cyphral Distich cipher by Scottish writer Thomas Urquhart in 44 minutes using 174,000 tokens. The cipher consists of two 32-digit lines and reveals a political text calling for support of Charles II. The result was reported by third-party model testing firm Vals AI.

Why it matters

This demonstrates that modern large language models (LLMs) can break centuries-old, pattern-based historical ciphers that have long frustrated human cryptographers. It blurs the line between 'impossible' puzzles and routine inference tasks. It also highlights Anthropic's effort to make frontier models both more capable and significantly cheaper, having reduced cache read costs by 75%. This development could accelerate real-world applications in linguistics, code-breaking, and document restoration.

Highlights

  • The model mapped each digit to the word position in Urquhart's previous 32 'wishes' and then took the first letter of each word. This was a simple substitution key that had gone unnoticed for 373 years.
  • 44 minutes and 174,000 tokens represent a modest compute budget, indicating that the task relied on pattern matching rather than brute-force search.
  • The result was announced by Vals AI, not Anthropic, meaning independent verification of the full methodology remains limited.

Critical take

The headline claiming the cipher was 'unsolved for 373 years' exaggerates the achievement. The cipher relied on a key already present in the source text (the previous 32 wishes), making it a missed clue rather than a truly unsolvable puzzle. The article also blends product marketing elements (pricing tiers, benchmark scores, EU AI Act-compliant watermarking) with the cipher story. The source is a French developer forum, not a primary or peer-reviewed account, so technical details regarding Fable 5.1's chain of thought remain unverified.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue