Codex now encrypts prompts and runs inference on the ciphertext
OpenAI's Codex has started encrypting user prompts and using that directly for model , rather than processing plain text. Previously prompts moved through the un; this change reduces the window during which prompt content is exposed on the server side. The move appears aimed at strengthening data protection, given that Codex is often used with proprietary or sensitive code.
Key points
- Codex now encrypts prompts before using them for
- Reduces the window where prompt content is exposed in plain text
- Aimed at protecting sensitive or proprietary code used with the tool