OpenAI's Codex Now Encrypts Instructions Between AI Agents — Leaving Developers Blind to Internal Delegation
OpenAI has introduced encrypted communication between AI agents in Codex, preventing developers from seeing how agents delegate tasks internally.
OpenAI Codex Encrypts Agent-to-Agent Communication
OpenAI has introduced encrypted communication between AI agents in Codex, preventing developers from seeing how agents delegate tasks internally. This feature encrypts inter-agent instructions, making multi-agent workflows opaque.
Affected Models
The encryption is mandatory for GPT-5.6 Sol and Terra models, while Luna remains open. This means premium users on GPT-5.6 cannot see agent handoffs.
Why Is OpenAI Doing This?
Possible reasons include protecting proprietary distillation processes, preventing prompt harvesting from multi-agent chains, and privacy guarantees for enterprise customers who don't want their internal logic exposed.
Developer Concerns
Developers report failed agent handoffs being harder to debug. The opaque nature of encrypted agent communication makes troubleshooting multi-agent workflows significantly more difficult.