OpenAI experienced a widespread login outage across its services on Wednesday, concurrent with patching a critical Codex bug that deleted user files and announcing zero data retention for frontier models.
OpenAI suffered a widespread outage on Wednesday evening that completely blocked logins and new account signups across chatgpt.com, its Codex coding platform, and roughly 12 API endpoints serving users in the United States and Europe. Those attempting to access the services encountered infinite loading animations and received explicit warnings about too many concurrent requests when they tried to send messages, before the company acknowledged the ongoing incident.[1]
During the same period, the company was forced to deploy an emergency fix for a severe bug in Codex running on the GPT-5.6 Sol model, where a specific command meant to clean up temporary files instead wiped out users' actual data. The destructive problem stemmed from the model improperly misusing system variables such as $HOME, thereby directing the automated deletion process to the real user home directory rather than the intended temporary folder.[2]
Beyond the immediate technical disruptions, the company also updated its core privacy controls, introducing a zero data retention policy for its frontier models to offer enterprise customers abuse monitoring without keeping their proprietary data. TechCrunch reported that under the new rules, when an abuse signal fires, the customer retains full control and decides whether to share data for review, establishing a clear contrast with Anthropic's mandatory 30-day retention window for its own equivalent models.[3]
