
What happened
Anthropic's Claude Fable 5.1 deciphered the Cyphral Distich, a 1653 puzzle by Sir Thomas Urquhart, in 44 minutes without human help. The solution spells a royalist message supporting King Charles II.
Why it matters
The puzzle was considered unsolved, and other frontier models failed to produce a verifiable solution. Fable 5.1 succeeded through systematic trial and error, not advanced cryptanalysis.
What to watch
The test is whether the 44-minute solve reflects a genuine leap in AI reasoning or merely persistence, since humans could have solved it too. The model’s ability to flag the 1653 puzzle as promising is what sets it apart from other frontier models that failed.
Summaries like this, every morning.
The Cyphral Distich consists of two lines of 32 numbers each, pointing to words in Urquhart's publication. Researchers had spent months testing other models on unsolved puzzles, but none yielded results. Fable 5.1's success highlights the role of persistence in AI problem-solving, potentially inspiring more attempts at historical puzzles.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
CVS Health appointed Jack Hanlon, who previously held senior data and analytics roles at Meta and Reddit, as s…

Salesforce customers are bracing for new fees to use Anthropic's Claude inside Salesforce apps, per The Inform…

Uber kept its AI spending flat while increasing usage 9.4 times, as reported in an August 2026 AXIOS exclusive…

Mastercard used the Money 20/20 conference to highlight AI agents, agentic commerce and stablecoins as key the…

Banks are moving to restrict AI agent access to customer accounts, a shift framed by USA Today as a 'window of…

AWS published a hands-on guide showing Multi-Turn Reinforcement Learning on Amazon SageMaker AI took GPT-OSS-2…
