Claude outage: Anthropic fixes worldwide error spike

Anthropic’s Claude service went offline for some users early Wednesday AEST, disrupting chats and developer work while the company dealt with elevated error rates across multiple models. Anthropic’s incident page says the trouble began at 12.08am AEST and was marked resolved at 2.44am.

The status log gives a tight, same-night timeline. Anthropic said it opened an investigation at 12.19am AEST, identified the issue six minutes later and had a fix in place by 12.53am. By the next update, it said success rates had broadly returned to normal across all models, though the company did not say what caused the failure.

“We have seen error rates broadly return to normal success levels across all models.”
Anthropic, Claude Status

The outage was short, but users noticed it. TechRadar reported that some people could not load chats or get prompts to complete. That kind of failure is more than an annoyance when Claude is being used to draft text, summarise documents or support software work; a stalled request can push teams back to manual steps or interrupt coding sessions built around the model being available on demand.

For Anthropic, the problem was partly about timing. Claude is no longer pitched only as a chatbot for occasional use. The company is selling it more directly into workplace and developer workflows, where reliability is judged closer to cloud software than a consumer app. Customers will want to know not only whether Claude came back online, but how often these interruptions happen and how clearly Anthropic explains them afterwards.

Why uptime matters

Public status pages rarely spell out the root cause during an incident, and Anthropic had not published a postmortem at the time of writing. Still, the update trail shows the company following a familiar cloud-software playbook: acknowledge the fault, identify it, deploy a fix and wait for success rates to recover before closing the incident.

The incident history page also gives customers a record to weigh against their own tolerance for downtime. A failed prompt in casual use is irritating. In an enterprise setting, the same failure can hold up document review, support drafting or coding work that depends on a model being reachable at the right moment. For an AI company trying to deepen workplace adoption, public reliability is part of the product.

That shift from novelty to infrastructure is why outages now matter more for AI assistants. The same model family that can summarise notes or draft emails is also being pulled into routine business processes. Buyers comparing AI tools will still look at model quality and features, but visible uptime and transparent incident handling will shape trust too.

Anthropic said service had returned to normal by the end of the incident window, and the public status page remained the clearest account of the event at publication time. The harder test is whether workplace users keep treating Claude as dependable enough for everyday work after another visible interruption.