A 370-Year Mystery Unraveled in 44 Minutes
AI evaluation and benchmarking startup Vals AI has revealed a striking demonstration of frontier reasoning capabilities: Anthropic’s Claude Fable 5.1 successfully decrypted the 'Cyphral Distich', a historical cryptogram that had remained unsolved for more than 370 years.
Announced over the transition of August 31 and September 1, 2026, the demonstration showcased the model operating autonomously for 44 minutes. Across that duration, the model consumed roughly 174,000 to 176,000 internal reasoning tokens in a single sustained inference loop, arriving at the deciphered text without external prompting or intermediate human steering.
The Cyphral Distich was originally published in 1653 by Scottish polymath and translator Sir Thomas Urquhart at the conclusion of his treatise 'Logopandecteision'. Consisting of exactly 64 numbers structured across two lines of 32 integers each, the cryptogram had resisted decades of specialized manual study and analytical attempts by historical cryptographers.
The Decryption Mechanism: In-Text Positional Mapping
Vals AI reported that Claude Fable 5.1 determined the cryptographic key resided within the surrounding source text itself. The model hypothesized and confirmed a structured mapping: the i-th number in the distich corresponds to the i-th 'Proquiritation'—a series of distinct petitions and pleas preceding the cryptogram in Urquhart's treatise.
By treating each individual number as a positional word index within its corresponding Proquiritation, the model extracted the initial letter of the targeted words. Stringing these extracted initials together produced a coherent 17th-century Royalist rhyming verse: 'O God, uphold King Charles the Second and make him the supreme ruler of this land', preserving a classic rhyming distich structured around 'and' and 'land'.
Rather than performing simple substitution sweeps, the model maintained an internal hypothesis evaluation loop, systematically checking structural relationships across multi-page early modern English prose until the positional indexing rule converged into grammatically and contextually sound plaintext.
Empirical Evidence and Cryptographic Verification
While the decoded text aligns seamlessly with historical sentiments surrounding the Stuart Restoration, cryptographic and literary academics are maintaining rigorous independent review. A key area of verification involves cross-referencing variant 1653 physical printings, where 17th-century typesetting ambiguities—such as Latin ligature conventions like 'hinc inde'—can alter exact positional word counts.
Because physical variations in surviving folios can introduce discrepancies into strict index-based ciphers, academic cryptographers continue to review the reconciliation across all primary printings before cataloging the puzzle as fully resolved in historical literature.
Furthermore, business leaders should note that the resolution was demonstrated and published by Vals AI, a specialist testing startup. While the algorithmic milestone is verified as an uninterrupted run, benchmark demonstrations of this caliber evaluate ceiling capabilities rather than out-of-the-box reliability across arbitrary unstructured enterprise data.
Practitioner Reactions: True Persistence Versus Computational Brute Force
Across engineering circles and the wider AI development community, the demonstration provoked immediate debate regarding the frontier of automated cognition. Skeptics questioned whether burning through more than 170,000 reasoning tokens across a 44-minute autonomous loop constitutes genuine deduction, or simply an expansive, compute-heavy brute-force sweep.
Conversely, many practitioners pushed back against the brute-force label. They highlighted that the combinatorial search space of early modern linguistic ciphers is virtually infinite; testing every potential structural relationship would take unfeasible amounts of compute. In their view, the model exhibited deep algorithmic persistence—formulating, rejecting, and refining complex hypotheses where previous human attempts stalled primarily due to limited human attention spans rather than intellectual insufficiency.
Other technical observers raised broader questions about the future of discovery. If autonomous agents can tackle historical cold cases and untangled textual structures without fatigue, developers anticipate a structural pivot where human researchers transition from exhaustive manual deduction to framing high-value hypotheses and auditing agentic outputs.
Strategic Implications for Enterprises in Thailand
For enterprise executives and technology leaders in Thailand, the significance of this milestone extends far beyond cryptographic history. It marks a decisive technological shift from instantaneous conversational completion toward deep, long-horizon autonomous problem-solving.
Thai enterprises in heavily regulated sectors—such as banking, insurance, legal advisory, and energy—frequently grapple with extensive documentation, historical contracts, regulatory filings, and complex cross-referenced corporate ledgers. The demonstrated ability to maintain coherent context and hypothesis testing across 170,000 reasoning tokens indicates that autonomous agents can increasingly handle high-stakes compliance audits, automated forensic accounting, and merger-and-acquisition due diligence without hallucinating under length pressure.
Nevertheless, Thai enterprise architectures must account for latency and unit economics. Deploying deep reasoning models for prolonged autonomous runs entails significant computational cost and multi-minute response delays. Organizations should implement tiered orchestration architectures: utilizing fast, cost-efficient models for standard document routing while reserving deep-reasoning engines like Claude Fable for edge-case investigations, legal discrepancy resolution, and high-value forensic analysis.
The decryption showcases autonomous long-horizon reasoning, signaling that frontier models can maintain coherent hypothesis-testing loops for complex audit, discovery, and analytical tasks that previously exhausted human attention spans.