OpenAI's Codex Begins Encrypting Prompts, Employs Ciphertext for Inference

OpenAI's Codex has added a new privacy layer to its workflow. The system now encrypts user prompts before any processing occurs. Encryption transforms the original text into ciphertext. Inference is carried out directly on the encrypted representation. This method is intended to keep prompt data hidden from internal logs. The change was documented in a public GitHub issue. Developers can examine the issue for implementation details. The approach may set a precedent for secure AI inference.