OpenAI AI Updates: May 9, 2026

1. OpenAI documents how it runs Codex safely internally

OpenAI. OpenAI published an account of the controls it uses to operate Codex coding agents internally, leaning on sandboxing, approval gates, network policies, and agent-native telemetry to keep autonomous code execution contained. The post is positioned as a reference for enterprises trying to ramp coding agents without exposing source trees, secrets, or production systems to unconstrained agent actions. Source

2. SoftBank trims OpenAI-backed loan from $10B to $6B

OpenAI. SoftBank reportedly cut an OpenAI-collateralized credit facility from $10 billion to about $6 billion after lenders balked at marking unlisted AI shares at the valuations SoftBank wanted. The downsize is a market signal on how far syndicated lenders will stretch on private AI paper, even with OpenAI as the underlying asset. Source