Codex logging bug can fill local SSDs with terabytes

OpenAI Codex has a verified logging bug that writes each request and response to a flat file without size limits, allowing the log to grow by terabytes and fill local SSDs [GitHub Issue]. The issue was opened on June 22, 2026 and has attracted 148 up‑votes from the engineering community, indicating widespread concern.

Current status

The bug is tracked in the OpenAI Codex repository; no official patch or workaround has been published yet. The issue description notes that the logger creates a new file per API call and never rotates or truncates it, leading to rapid disk consumption. OpenAI has not commented on a timeline for a fix.

Why it matters

Uncontrolled log growth can exhaust SSD capacity, causing services to crash or become unresponsive. In production deployments, the storage pressure forces engineers to implement temporary mitigations—such as disabling the logger, redirecting output to a network share, or manually pruning files—to keep systems operational. The bug also raises questions about the reliability of Codex in long‑running workloads, where silent storage exhaustion could go unnoticed until failure.

Mitigation guidance

Until a fix lands, teams should:

• Enable filesystem monitoring to alert when disk usage spikes.
• Configure the logger to write to a dedicated, sized volume or to a remote storage endpoint.
• Consider disabling verbose logging in production environments.

Developers are encouraged to follow the GitHub issue for updates and to contribute test cases that reproduce the excessive logging behavior.

Poll: What is your primary concern when integrating AI models like OpenAI Codex?

• Storage capacity
• Model reliability
• Data security
• Development time