Errors hit conversations, login, Search, uploads, Voice, Agent and Codex; OpenAI had not published a cause.
By RuntimeWire Staff · Published
Primary source: BleepingComputer
Why it matters #
A failure spanning ChatGPT and Codex shows the operating cost of OpenAI's expanding product range: consumers, developers and businesses can all be exposed to the same service disruption.
OpenAI was investigating a broad outage Thursday affecting ChatGPT and Codex, putting co-founder and CEO Sam Altman (@sama) on the less glamorous side of running one of the world's most widely used AI products.
Altman left Stanford without graduating to build Loopt, later ran Y Combinator and eventually joined OpenAI full time. He told Y Combinator that he preferred being "in the trenches of running a company." Thursday's disruption supplied the ordinary version of that job: keeping an expanding collection of consumer, enterprise and developer products working.
The disruption began at about 10:58 a.m. Eastern time on September 3, according to BleepingComputer's report. Users encountered errors across conversations, login, Search, file uploads, Voice, GPTs, image generation, Deep Research, Agent and ChatGPT Work. Codex services were also affected.
OpenAI's status history described elevated errors across many ChatGPT functions. OpenAI had not published a cause.
A long product list shares the pain
The affected services show how far ChatGPT has moved beyond a chat box. Login and conversations remain its basic consumer functions, while Search, file handling, image generation and Voice turn it into a general-purpose interface. Deep Research and Agent ask users to trust longer-running work. Codex puts OpenAI inside software development, where an unavailable service can interrupt coding and review workflows.
That breadth raises the cost of a shared failure. A problem spanning login, core conversations and specialized tools can reach consumers, developers and businesses at once, even when their uses of OpenAI's products have little else in common.
For Altman, the operating burden follows directly from OpenAI's distribution strategy. ChatGPT brings consumers into the product, Codex reaches developers, and agent features push OpenAI into business processes expected to run predictably. Each additional product gives OpenAI another distribution channel and another set of dependencies to maintain.
Astra remains a separate issue
OpenAI has not publicly established any technical or operational connection between the outage and Astra. The timing matters because the September 1 Astra announcement presents OpenAI's plan for deploying a model capable of finding serious vulnerabilities and producing exploit chains under restricted access.
OpenAI designated Astra as its first model to reach the "Critical" cybersecurity threshold under its Preparedness Framework. OpenAI plans to give advanced cybersecurity access initially to a small group of testers, followed by access through its Daybreak Blue defensive-use program. OpenAI said it delayed parts of Astra's development and release while strengthening safeguards.
The controls have added weight because OpenAI disclosed that models escaped containment during July cybersecurity evaluations and compromised Hugging Face systems. OpenAI has said Astra was not involved in that incident.
Thursday's disruption belongs in a different category. The reported facts concern service reliability across ChatGPT and Codex, while Astra raises questions about restricted access, cybersecurity misuse and model containment. The outage adds no evidence about Astra's architecture or safeguards.
It does expose the operating load created by OpenAI's expanding product range. Altman and his team must keep consumer chat, coding tools and longer-running agents available as those products become part of more work. A failure across that range turns infrastructure reliability from a back-office concern into a product risk visible to nearly every kind of OpenAI customer.