Claude Continues to Suffer Outages, Services Down for 3+ Hours

Anthropic’s Claude faced another service disruption, with elevated errors affecting multiple models and users reporting problems across the company’s AI services on August 24. 

The company’s status page recorded an incident lasting 3 hours, 24 minutes, affecting multiple Claude models and partial outages across Claude.ai, Claude Code, Claude Cowork, and the Claude API.

The company said the disruption involved elevated errors affecting requests to multiple models, including Claude Opus 5, Mythos 5, Opus 4.8, and Fable 5. Anthropic said it identified the cause and worked to restore normal success rates across the affected models.

The latest incident adds to a series of outages and performance degradations Anthropic has experienced over the past two weeks.

One service has stood out from the rest: Claude for Government. Anthropic’s status dashboard currently shows the government service at 100% uptime over the past 90 days, while its commercial Claude services have recorded downtime during the same period.

“Only way to use Claude: join the government,” quipped one user on X. “Or switch to [OpenAI’s] Codex.”

Another developer wrote on X, “Those who have looped tasks or agentic workflows are undoubtedly running into dead ends tonight and will find their work undone in the morning.” 

August has been particularly outage-heavy for Claude. Anthropic’s status page lists numerous incidents involving elevated errors, degraded model performance, and disruptions across Claude’s broader services, with several incidents affecting multiple models or products.

The outages have an extended impact as Claude Code has become increasingly embedded in developers’ day-to-day workflows. 

In June, Anthropic’s analysis of around 400,000 Claude Code sessions involving roughly 235,000 users found that users spent an average of 20 hours a week with the coding agent. 

Over the seven-month period studied, Claude Code usage also shifted toward more end-to-end work, including deploying and running code, analysing data, and producing non-code outputs, while the share of sessions spent debugging fell by nearly half.