A week of ChatGPT Pro quota burned in 36 hours as GPT-6 fails to factor a 512-bit RSA key

matthew_d_green · x · 2026-10-16

Cryptographer Matthew Green shares a notable experiment: someone burned a week's quota of max-tier ChatGPT Pro in 36 hours, driving GPT-6 Astra Ultra to derive an L(1/4) factoring algorithm and factor a 512-bit RSA key.

The model wrote extensive Lean proofs and proved intermediate impossibility results, but made effectively no progress.

The experimenter had to prompt carefully to evade oppressive security guardrails (Daybreak Blue doesn't apply to GPT-6 Astra), calling this another empirical point that AI won't magically produce a fast factoring algorithm.

Original post →

More from Models

Models channel →