Claude Went Dark for an Hour: Inside Anthropic's Second Major Outage in Seven Days

On Tuesday morning, Claude stopped working. Not just the chatbot — the web app, the desktop and mobile apps, the coding tool, the developer API, all of it. For roughly an hour, one of the most widely used AI systems in the world was effectively unreachable, and Anthropic's own status page read like a real-time incident log.

The trouble began at 10:21 a.m. Eastern, when Anthropic started investigating what it described as "elevated error rates" across claude.ai, Claude Code, Claude Cowork and the Claude API. About ninety minutes later, the company said everything was back to normal. In between came failed requests, broken logins, unsaved conversations, and a warning to users that read, simply: if you're signed in, please don't sign out.

It was the second significant Claude outage in seven days — and, according to incident trackers, the 13th entry on Anthropic's status page this month.

Photo: Anthropic CEO Dario Amodei at TechCrunch Disrupt 2023. Credit: TechCrunch, via Wikimedia Commons (CC BY 2.0).

The outage, minute by minute

Anthropic's status page gives the clearest account of what happened, and the sequence shows a morning that kept getting worse before it got better. All times below are Eastern, with UTC in parentheses.

10:21 a.m. ET (14:21 UTC): Anthropic opens an investigation into elevated error rates affecting claude.ai — including the desktop and mobile apps — Claude Code and Claude Cowork. Some requests to the Claude API are also returning errors. The company tells users they may see failed requests, errors loading or sending conversations, or be asked to sign in again, and that retrying may succeed.

10:28 a.m. ET (14:28 UTC): The affected-service list grows to include claude.ai, platform.claude.com (the Console), the Claude API, Claude Code and Claude Cowork, all under degraded performance.

10:41 a.m. ET (14:41 UTC): A mitigation is applied and error rates drop substantially — but some sign-in attempts, account actions, and Claude Code and Cowork sessions are still failing. Engineers keep working toward a full fix.

11:00 a.m. ET (15:00 UTC): The plot twists. Anthropic says the errors that began at 10:00 a.m. ET were mitigated at 10:36 a.m., with existing conversations mostly working again — but a second issue is now blocking many users from signing in. Single sign-on and Sign in with Apple are unavailable, along with new chats, voice conversations, Claude Code and Cowork sessions, purchases and file uploads. Hence the memorable advisory: "If you're signed in, please don't sign out."

11:11 a.m. ET (15:11 UTC): Most services recover as of 10:59 a.m. ET. Signing in, starting new chats, voice conversations, Code and Cowork sessions, purchases and file uploads work again. But there is a sting in the update: "Some messages sent between 14:00 and 14:59 UTC may not have been saved."

11:30 a.m. ET (15:30 UTC): "All services have been operating normally since 14:59 UTC," the status page says. The company continues to monitor.

12:27 p.m. ET (16:27 UTC): The incident is formally marked resolved, with impact recorded from 7:00 to 7:59 a.m. Pacific (14:00–14:59 UTC) — essentially a one-hour window.

Everything broke at once

What made Tuesday's outage notable was its breadth. Anthropic classified claude.ai, Claude Code and Claude Cowork as partial outages, while the Claude Console and the API were listed under degraded performance, according to NDTV Profit's summary of the status page.

For users, the disruption took several forms: unsuccessful requests, trouble loading or sending conversations, and prompts asking them to sign in again. Anthropic's advice throughout was pragmatic — keep retrying, because some attempts would go through.

The scale was large enough to register clearly on Downdetector. NDTV Profit reported more than 11,113 reports by 7:47 a.m. Pacific, with Claude Chat accounting for most of them. Channel NewsAsia, relaying Reuters data, said US reports topped 10,000 around 10:25 a.m. ET before falling below 100, with a smaller Singapore spike of about 360 reports. Moneycontrol's chart showed an even sharper peak of 18,335 reports. Downdetector's figures are based on user-submitted reports, as CNA noted, so the true number of affected users may differ — but the spike shape tells the same story everywhere: a sudden, widespread failure, then a fast recovery.

Photo: Server racks in a data center — the infrastructure tier behind services like Claude. Credit: Carl Lender, via Wikimedia Commons (CC BY 2.0).

Spotify went down at the same time

Claude was not the only service having a bad morning. Spotify suffered its own disruption beginning around 9:30 a.m. ET, with Downdetector reports peaking above 19,000 in the US before dropping back near 1,000, according to CNA. Spotify acknowledged "some issues" in a post on X and said it was investigating.

There is no evidence the two outages were linked — as TechDefused put it, the simultaneous failures appear to be coincidence, with neither company explaining what went wrong on its own infrastructure. Still, the pairing made for a rough morning for anyone whose workday soundtrack and AI assistant both live in the cloud.

The second outage in seven days

The September 29 failure came just one week after another major Claude disruption. On September 22, an incident titled "Elevated errors for multiple models" ran from 00:57 to 02:35 UTC, with impact recorded from 00:50 to 02:10 UTC — about an hour and a half of elevated errors across Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5, according to Unite.ai's review of the status history. That time, Anthropic said it had identified the cause and restored normal success rates. Downdetector logged more than 1,600 reports during that earlier episode.

This time the damage was broader: not just models, but the full stack — consumer apps, developer tools, the API and the account system. And Tuesday's incident was the 13th logged on Anthropic's status page this month. Partial outages also hit Claude on September 11 and 15, and on September 3 Claude went down at the same time as ChatGPT and Grok in a wider AI disruption, per 9to5Google.

Anthropic's own numbers put September uptime at 99.70%, up from 99.59% in August and 99.01% in July, according to Unite.ai. That sounds reassuring until you do the arithmetic: it still represents hours of downtime in a month — for a service that developers, businesses and students increasingly treat as load-bearing infrastructure.

The questions Anthropic didn't answer

The most conspicuous feature of Tuesday's incident report is what it leaves out: a cause. The status updates walk through the timeline in detail — the mitigation at 10:36 a.m. ET, the second failure, the recovery — but never say what actually broke. Anthropic has not disclosed a root cause.

The timing will invite speculation regardless. On Monday, September 28 — one day before the outage — Claude Sonnet 5.5 became the default Sonnet model on Anthropic's API, a launch TechPulse USA covered earlier this week. That is reported context, not a confirmed cause, and nothing in Anthropic's updates connects the two. But when a company swaps the default model on its API on Monday and spends Tuesday morning fighting a full-stack outage, users are going to ask.

The deeper issue is dependency. When sign-in is dead for an hour, it's not just a chatbot that disappears: it's coding sessions in Claude Code, automated workflows built on the API, and conversations that turn out not to have been saved. Enterprise appetite for AI keeps accelerating — OpenAI alone is approaching a $70 billion revenue run rate on enterprise demand — and every hour of ambiguity about what broke is an hour when a business customer's faith in its AI vendor erodes a little.

Anthropic is reportedly headed toward one of the largest IPOs in tech history. Reliability is the currency such companies spend years accumulating and can lose in a morning. Two outages in a week, with no public explanation for the second, is the kind of record that investors, enterprise buyers and competitors all notice. For now, Claude is back — but the silence on what knocked it over is the part of the story that hasn't been resolved.

Sources: This article draws on Anthropic's status page at status.claude.com (primary), AI Weekly, Channel NewsAsia (relaying Reuters), NDTV Profit, and Moneycontrol for the September 29 outage; Unite.ai and 9to5Google for the September 22 incident and monthly incident history.