GPT-6 Astra Cracks 217-Year Cipher
OpenAI’s GPT-6 Astra reportedly decoded a 1809 Napoleonic military cipher from a single low-resolution scan in roughly six hours, combining transcription, web research, partial-key discovery, and cryptanalysis. The result was reported by SentinelOne engineer Carter Church and covered by Euronews.
The impressive part is the end-to-end research loop, not merely recognizing symbols: Astra connected scattered historical evidence to a mechanically testable plaintext. It is a compelling capability demo, though independent historians still need to validate the reconstruction.
- –Astra identified and applied a previously unlinked partial cipher table covering roughly one-third of the symbols.
- –The workflow shows why multimodal reasoning, web search, tool use, and long-running context matter for research agents.
- –Developers can access Astra through the API with a 1.05-million-token context window and support for web search, code execution, computer use, and MCP.
- –The demonstration is an individual report rather than a controlled benchmark, so it should not be treated as proof that Astra can reliably solve arbitrary historical ciphers.
- –Verifiable outputs remain essential: fluent plaintext is not enough without historical consistency and cryptographic validation.
DISCOVERED
1h ago
2026-10-04
PUBLISHED
1h ago
2026-10-04
RELEVANCE
AUTHOR
AI Revolution