We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·21d ago·5 min read
Codex starts encrypting prompts, uses ciphertext for inference instead
TL;DR
Codex has implemented encryption for prompts, utilizing ciphertext during inference to enhance security. This change specifically impacts the MultiAgentV2 framework, affecting message handling functions.
✦ Why It Matters
Engineers should update their MultiAgentV2 implementations to utilize the new encryption features for enhanced security.
Key Takeaways
- Review the impact of encryption on your debugging processes.
- Advocate for an audit field in Codex to retain task visibility.
- Test the latest Codex version to understand changes in message handling.
- Monitor updates for potential fixes to the current audit trail issues.
How It Works
Codex encrypts messages in the MultiAgentV2 framework, storing only the encrypted content while leaving the original message fields empty. This design choice prioritizes privacy but sacrifices the ability to audit and debug effectively.