OpenAI's GPT-6 Astra model cracked the Enigma code in 2 days

Summary

ZDNET France reports that OpenAI's GPT-6 Astra model autonomously decrypted a previously unsolvable World War II Enigma message within two days. The model navigated archives, identified an original typographical error, and wrote its own Python/C++ simulation tools. The article frames this event as a proof of concept for autonomous AI research agents applicable to corporate problems.

Why it matters

If this claim is accurate, the event can be read as a significant step. AI systems are becoming capable of independently exploring complex, unstructured information domains and generating their own tools without step-by-step human guidance. From a corporate perspective, this capability could change how teams solve problems in areas such as hypothesis generation, code execution, multi-source synthesis, supply chain analytics, security auditing, and process optimization.

Highlights

• GPT-6 Astra is reported to have detected a historical typographical error and built its own Python/C++ simulators to break the cipher.
• Human experts had spent weeks on the same message; the AI completed it in two days.
• The article positions this event as a template for autonomous agents handling unstructured corporate problems.

Critical take

The article reads more like promotional copy than journalism. It does not specifically name the decrypted Enigma message and offers no independent verification. The leap from solving a single historical puzzle to a general claim about corporate-level autonomous reasoning is also exaggerated. Additionally, the phrase 'two days' is ambiguous. Continuous computation time and a two-day project window involving human checkpoints are very different things; the article fails to make this distinction.