OpenAI has introduced encryption for instructions passed between AI agents within its Codex system, a change implemented since early June. This update prevents developers from monitoring how tasks are delegated internally among AI subagents, according to The Decoder.
The encryption feature is mandatory for the GPT-5.6 Sol and GPT-5.6 Terra variants, marking a significant shift in transparency around AI task management. By encrypting these communications, OpenAI aims to secure internal processes but limits external visibility into AI decision flows.
For Japanese markets, where AI-driven trading and automation are increasingly prevalent, this move raises questions about transparency and control in AI-powered systems. Understanding how AI agents allocate tasks internally can be critical for risk management in FX, crypto, and equities sectors.