OpenAI's Codex coding tool has been writing so much data to a hidden log file that some users fear it could wear out their drives early. One developer watched their main drive take about 37 TB of writes in roughly 21 days, and traced almost all of it to Codex.
That works out to around 640 TB a year. Many consumer drives are rated for only about 600 TB of total writes before the maker calls them worn out, so Codex could burn through a drive's warranty in under a year.
Codex is OpenAI's command-line coding helper. It keeps a running feedback log in a small local file on your computer. That file, plus two companion files, was the thing hammering the disk. A developer flagged it on June 14 in a GitHub issue after checks pointed to the Codex logs as the top constant disk writer.
How a small file does so much damage
The same tiny log got written and erased over and over. It held only about half a million entries at a time, yet its internal counter had already handed out more than 5.5 billion entry ID numbers. The biggest culprit was Codex saving every piece of the AI model's streamed replies, which made up most of the data.
A fix is already out
OpenAI merged two fixes on June 22. One stops Codex from logging every piece of the streamed replies, and the other filters out the noisiest sources. The developer estimated the changes would cut about 85 percent of the logging in their setup, then closed the issue.
Update to the newest version, but keep watching. Other reports say the heavy writing kept happening on the Windows desktop app, so the fix may not catch every setup.
The public issue shows no sign that any data left the user's computer. Still, the logs may include your conversations, and OpenAI has not yet answered the privacy question this raises.
