Major AI services hit simultaneous outage Thursday morning
On Thursday morning, Anthropic’s Claude models saw elevated error rates starting at 9:23 a.m. ET, OpenAI’s ChatGPT and Codex degraded from 10:43 a.m. ET, xAI’s Grok showed a spike in user reports from under ten to 1,365 by 9:45 a.m. ET, and Google’s Gemini API was flagged as likely down between 10:45 a.m. and 11:15 a.m. ET.
Short answer: On Thursday morning, Anthropic’s Claude models saw elevated error rates starting at 9:23 a.m. ET, OpenAI’s ChatGPT and Codex degraded from 10:43 a.m. ET, xAI’s Grok showed a spike in user reports from under ten to 1,365 by 9:45 a.m. ET, and Google’s Gemini API was flagged as likely down between 10:45 a.m. and 11:15 a.m. ET.
Major AI services simultaneous outage Thursday morning
Thursday morning kicked off with a surprising hiccup across several big-name AI services, all acting up at nearly the same moment. Anthropic was the first to acknowledge trouble, saying its Claude models were seeing elevated error rates beginning at 9:23 a.m. Eastern. The company traced the problem within about fifteen minutes, deployed a fix, and marked the incident resolved by 12:16 p.m. A separate note later indicated a brief spike in errors for the Claude Sonnet 5 model just after noon.
OpenAI ChatGPT and xAI Grok outage details
OpenAI followed with its own alert, noting that ChatGPT and its Codex counterpart were suffering from degraded performance starting at 10:43 a.m. Eastern. Engineers put a mitigation in place roughly half an hour later, and the service was reported as restored by 12:55 p.m.
Meanwhile, xAI’s Grok showed a user-facing warning that the model was experiencing issues. DownDetector data reflected a sharp rise in reports, climbing from fewer than ten before 9 a.m. to 1,365 by 9:45 a.m. By early afternoon the number had fallen to 273, indicating the problem was easing but not yet fully cleared.
Google Gemini outage signs and cloud provider status
Google did not issue a public statement about its Gemini model, yet independent monitors painted a similar picture. DownDetector logged a jump from 23 reports around 10:30 a.m. to 412 just after 11 a.m. StatusGator, an API-watching service, flagged a likely outage for the Gemini API between 10:45 a.m. and 11:15 a.m., after which the status returned to normal.
Other major cloud providers appeared unaffected. Amazon Web Services, Microsoft Azure, and Cloudflare did not report any widespread problems, although their own DownDetector traces showed modest bumps in user reports during the same morning window.
Why multiple AI services failed at once and what it means
The coincidence of four frontier AI services stumbling within a few hours is unusual. Anthropic’s own metrics show its Claude family maintained 99.4 % uptime over the past ninety days, with the last comparable three-hour partial outage occurring on August 24. OpenAI reports ChatGPT averaged 99.63 % availability in the same period, while its Codex variant logged a perfect 100 % score. The only recent hiccup for OpenAI’s Work Mode was an extended latency episode on August 31.
For developers and businesses that rely on these APIs, the episode underscores the importance of building redundancy into AI-dependent workflows. Even when individual providers boast high reliability scores, simultaneous failures can disrupt applications that assume at least one service will remain online. Teams may want to review fallback strategies, such as caching responses, switching to alternative models, or queuing requests during outage windows. Monitoring tools that aggregate status across providers can also give earlier warning when multiple services start to show error spikes.
While the root causes behind each incident have not been disclosed in detail, the timing suggests a possible shared stress factor, perhaps related to underlying infrastructure or a sudden surge in demand. Until more information emerges, the safest approach for anyone building with AI is to treat these platforms as potentially fall
Frequently asked questions
At what time did Anthropic first notice elevated error rates for its Claude models on Thursday?
Anthropic first acknowledged trouble with its Claude models at 9:23 a.m. Eastern, when it observed elevated error rates beginning then.
How long did it take for OpenAI to restore ChatGPT and Codex after the degraded performance began?
OpenAI reported degraded performance starting at 10:43 a.m. Eastern, implemented a mitigation about half an hour later, and said the service was restored by 12:55 p.m.
What did DownDetector show for xAI’s Grok and Google’s Gemini during the outage window?
DownDetector recorded Grok reports rising from fewer than ten before 9 a.m. to 1,365 by 9:45 a.m., then falling to 273 by early afternoon; for Gemini, reports jumped from 23 around 10:30 a.m. to 412 just after 11 a.m.
Which major cloud providers reported no widespread problems during the simultaneous AI service outage?
Amazon Web Services, Microsoft Azure, and Cloudflare did not report any widespread problems, although their DownDetector traces showed modest bumps in user reports during the same morning window.
Source: Ars Technica
Related posts
IBM's ALTK-Evolve Tops Customer Service Benchmark With GPT-4.1
IBM Research's ALTK-Evolve method propelled a GPT-4.1-based customer-service agent to the top of the τ²-bench leaderboard without fine-tuning or human labels. The agent automatically writes its own playbook by reviewing past trajectories and distilling behavioral rules into the system prompt. On retail tasks, Pass^4 reliability jumped from 57.5% to 67.5%; on airline tasks, a lean 14-rule core lifted Pass^4 from 35% to 45%. The approach adds only context tokens, no training overhead.
Goodfire Researcher Claims Neural Networks Are Understandable
Goodfire researcher Nick Cammarata claims neural networks may be far more understandable than previously believed. After years of assuming models encode information in tangled "superposition," Cammarata re-examined a neuron firing for both cars and cats, finding it detects V-shaped geometric features. He argues early confusion created confirmation bias. Skeptics, including Stanford's Chris Potts, call the claim an unproven hunch. Goodfire plans to publish technical results applying the
CoreWeave Debt Crisis Rattles AI Infrastructure Sector
CoreWeave's bonds entered distressed territory, triggering a sector-wide selloff. The GPU cloud provider owes over $35 billion and lost $1 billion in the first half of 2024 despite a $103.7 billion revenue backlog. Investors fear a debt spiral as the company seeks investment-grade ratings to tap insurer capital while GPU oversupply looms. Revenue concentration among Microsoft, OpenAI, and Nvidia creates counterparty risk if AI demand falters.