Agents2h ago
OpenAI Codex Encrypts Agent-to-Agent Instructions
OpenAI's Codex now encrypts instructions passed from main agents to subagents. This change removes developer visibility into internal task delegation. Encryption is mandatory for the Sol and Terra GPT-5.6 variants. This shift prioritizes system security over transparency, forcing developers to trust the black-box orchestration of complex agentic workflows without granular audit logs.