“A Fix Has Been Deployed”: Four AI Chatbots Broke During L.A.’s Morning

Anthropic and OpenAI logged incidents that overlapped by about 90 minutes, and Google never publicly acknowledged Gemini's trouble.

Four of the most-used cloud AI models failed for users inside a single morning window on September 3, 2026. The window ran three and a half hours. Ars Technica assembled the overlap from the companies' own incident notes and two outage trackers, covering OpenAI, Anthropic, xAI and Google. The cause hasn't been made public.

In Los Angeles that window opened at 6:23 a.m. Pacific. It shut at 9:55 a.m. — the first three hours of the workday for anybody who drafts, codes or summarizes with one of these tools. The companies posted their notices in Eastern time, and we've converted every timestamp below to Pacific.

The timeline, in Pacific time

Anthropic posted first. Its status page opened a partial outage at 6:23 a.m., citing elevated errors on three Claude models:

• Claude Mythos 5.1
• Claude Fable 5.1
• Claude Opus 5

About 15 minutes later the company said it had identified the cause. At 9:16 a.m. it posted that “a fix has been deployed” and closed the incident. A separate, brief note covered elevated errors on Claude Sonnet 5 just after 9 a.m.

OpenAI's notice landed at 7:43 a.m.: elevated errors across ChatGPT and Codex, which the company called “degraded performance.” A mitigation went in a little more than half an hour later. The incident was marked resolved at 12:55 p.m. Eastern — 9:55 a.m. here.

Grok users, meanwhile, got an error screen saying the model was having problems and that xAI was working to bring service back. Just before 6 a.m. Pacific, fewer than 10 Grok reports had been filed on the outage tracker DownDetector, according to Ars Technica. Forty-five minutes later the outlet counted 1,365 of them. By the time it published, the count was down to 273.

Google hasn't publicly acknowledged any Gemini problem. DownDetector's Gemini count stood at 23 around 7:30 a.m., the outlet reported. It hit 412 just after 8 a.m. StatusGator, which monitors API status pages, logged what it calls a likely outage for the Gemini API between 7:45 and 8:15 a.m., after which the status went back to normal.

That's the timeline. It isn't a cause.

The overlap

Set the two official incidents side by side and they run together for about 90 minutes. OpenAI's opened at 7:43 a.m. Anthropic's closed at 9:16 a.m. StatusGator's half-hour Gemini window sits entirely inside that, and Grok's report spike came earlier, with its error screen still up when the story ran.

The baseline is what makes that unusual. Anthropic's status page reports 99.4 percent uptime for Claude over the past 90 days. OpenAI reports 99.63 percent for ChatGPT across the same period. Codex, on that same page, shows 100 percent.

Desk read

Amazon Web Services, Microsoft Azure and Cloudflare hadn't reported major problems as of press time, though report counts for all three ticked up on the same morning. That's the detail that invites a shared-cause theory, and it doesn't support one — a spike in user complaints measures how many people were annoyed, not what failed. Two of the four companies posted an incident. One showed an error screen. One said nothing.

The last comparable Claude event was August 24, 2026, also a partial outage of roughly three hours. ChatGPT's Work Mode logged hours of elevated latency on August 31.

Source: arstechnica.com, retrieved September 3, 2026.