OpenAI AI Updates: May 14, 2026

1. Codex on Windows ships with a security-hardened sandbox

OpenAI. OpenAI detailed the sandbox it built so the Codex coding agent can run safely on Windows, walking through the controlled file-access scheme and network restrictions that gate what the agent can touch on a developer’s machine. The post pitches the design as the foundation that finally lets Codex run with auto-approval on Windows the same way it does on macOS and Linux without expanding the blast radius if the model misbehaves. Worth reading alongside the Codex changelog if you’ve been holding off on auto-approval inside Windows dev environments. Source

2. OpenAI responds to the “Mini Shai-Hulud” TanStack npm supply chain attack

OpenAI. OpenAI published its response to the recent npm supply chain compromise that hit the TanStack ecosystem, dubbed “Mini Shai-Hulud,” outlining the detection path and the mitigations it pushed into the Codex toolchain. The most concrete user-facing item: macOS users must update their OpenAI applications by June 12, 2026 or risk losing functionality, since older signed binaries are being rotated out as part of the response. The post is a useful reference for teams now hardening their own coding-agent pipelines against package-registry attacks. Source

3. Federal court testimony renews scrutiny of OpenAI’s leadership

OpenAI. Sam Altman provided testimony in federal court proceedings concerning his trustworthiness as a business leader, and TechCrunch’s accompanying piece examines how the proceedings intersect with OpenAI’s governance and partner relationships. The angle is less a product story and more a temperature check on how regulators, investors, and customers are now weighing leadership risk at the largest US AI lab. Worth tracking for any team whose roadmap depends on the OpenAI stack. Source