The Pulse
Anthropic restores Claude services after 80-minute incident
Anthropic reported elevated errors affecting several Claude models and four Claude services on September 22, 2026. The company said success rates returned to normal before resolving the incident after 80 minutes.
AI.info Team ·
Anthropic restores Claude services after 80-minute incident
Anthropic resolved elevated errors affecting multiple Claude models on September 22, 2026, after an incident that affected Claude’s web application, API and developer tools.
The impact occurred from 5:50 p.m. Pacific Time to 7:10 p.m. Pacific Time, or 00:50 UTC to 02:10 UTC, according to Anthropic’s official incident record. The company posted the incident as resolved at 02:35 UTC. The total impact window lasted 80 minutes.
The affected services were claude.ai, the Claude API at api.anthropic.com, Claude Code and Claude Cowork. Anthropic’s status record identified those products as affected but did not provide separate error rates, request totals or regional information.
Incident affected three Claude model lines
Anthropic’s first update, posted at 00:57 UTC, said it was investigating elevated errors on requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5. That update did not identify a cause or specify how many requests were failing.
At 01:17 UTC, the company said it had identified the cause of the elevated errors and was working on a fix. The status record did not explain what caused the incident or describe the corrective action.
By 01:35 UTC, requests to Claude Fable 5 and 5.1 and Mythos 5 and 5.1 had returned to normal success rates. Anthropic said it was still working to resolve errors affecting Claude Opus 5. The update showed that recovery was not simultaneous across the affected model lines.
At 02:11 UTC, Anthropic reported that success rates had returned to normal across the affected models and said it was monitoring the recovery. The final status update, posted at 02:35 UTC, said the issue had been resolved and confirmed the start and end of the impact period.
Multiple entry points were affected
The incident was not limited to one model endpoint. The status record lists the Claude web application, the API, Claude Code and Claude Cowork among the affected systems. Users reaching the affected models through any of those services could have encountered errors during the 80-minute impact window.
Anthropic distinguished between the models reporting elevated errors and the services through which users accessed them. The models named in the incident updates were Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5. The later recovery update also referred to Claude Fable 5, Fable 5.1, Mythos 5 and Mythos 5.1, while stating that Opus 5 still had remaining errors at that time.
The sequence of updates records five stages: an initial investigation at 00:57 UTC, identification of the cause at 01:17 UTC, partial model recovery at 01:35 UTC, monitoring after broader recovery at 02:11 UTC and resolution at 02:35 UTC. Anthropic did not disclose the technical cause after saying it had identified one.
It also did not publish error percentages, affected regions, request counts or a detailed account of how the fault was corrected. The record does not say whether all users experienced the same symptoms or whether the errors were concentrated in a particular geography or access method.
Anthropic’s final record therefore establishes the timing of the disruption, the services affected and the models involved. It shows that some model success rates recovered before others, followed by a return to normal across the affected models before the incident was marked resolved.