Miles Brundage, a highly respected figure in artificial intelligence policy and former Head of Policy at OpenAI, has reported a significant malfunction in the latest build of a Codex-powered Mac application. On April 14, 2026, Brundage described an unprecedented failure mode, stating, > "Something seems to have broken in the latest Codex (Mac app) build. I have never seen this particular failure mode before (never-ending 'Reasoning [shimmering]' status with no actual substantive updates)." This issue suggests a severe disruption to the application's core functionality.
The reported bug manifests as a persistent "Reasoning [shimmering]" status, indicating that the AI is perpetually processing without delivering any tangible output or progress. This specific problem mirrors similar reports from users of GitHub Copilot, an AI pair programmer powered by OpenAI Codex, which often integrates into IDEs like Visual Studio Code on macOS. Developers frequently describe Copilot getting stuck on "Generating suggestions" with a shimmering or spinning indicator, particularly following recent software updates.
OpenAI Codex is a sophisticated AI model designed to translate natural language into code, significantly enhancing developer productivity. While OpenAI does not offer a standalone "Codex Mac app," its capabilities are widely accessible through integrations such as GitHub Copilot, which functions as an extension within various development environments. This suggests the issue Brundage encountered likely pertains to one of these integrated applications.
Brundage's background as the former Head of Policy at OpenAI lends considerable weight to his observations, highlighting potential challenges in the deployment and stability of advanced AI coding tools. Such persistent processing errors can severely impede developer workflows and underscore the complexities involved in maintaining robust performance for AI-driven applications. The recurrence of these "stuck" states, as noted by the wider developer community, indicates an ongoing area for improvement in AI code generation technologies.
