Feature Request: Response Authenticity & Prompt Confidentiality

ChatGPT
1

I’d like to suggest two much-needed security improvements for OpenAI and Anthropic.

  1. Response Authenticity with Digital Signatures

    Follow a PGP-like signing model:

  • OpenAI releases a verified public key.

  • Server-side sign every reply with sign(r.

  • Return the signature together with the output.

  • Clients can locally verify integrity to block man-in-the-middle tampering and prevent third-party AI relays from modifying or diluting official model responses.

  1. Prompt Confidentiality via Public Key Encryption

    Let clients encrypt user prompts and conversation context with OpenAI’s offical public key before transmission, so only OpenAI can decrypt it with the private key.

    To support third-party relay workflows, keep the system introduction optional in plaintext. This protects user privacy while still allowing proxies to apply custom rules. This is a critical feature request that can protect millions of GPT users relying on relays like OpenRouter, safeguarding their confidential data from leaks caused by untrusted intermediate providers.

These two features will bring cryptographic trust to AI responses and better protect user prompt privacy against intermediate forwarding services. It would be great if the official team could consider implementing this architecture.

2

This topic was automatically closed after 22 hours. New replies are no longer allowed.