Anthropic's Claude AI platform suffered a global outage today, leaving users in India and elsewhere unable to access the chatbot, its coding tool, and its mobile app for over three hours before services were restored.
What went wrong?
According to Anthropic's official status page, the company began investigating elevated errors on requests to Claude Mythos 5, Claude Fable 5, Claude Opus 5, and Claude Opus 4.8 at 5:06 am UTC (10:36 am IST) on August 24.
The company said it had identified the cause of the errors within about 20 minutes but continued working through the morning to stabilise all affected models. Anthropic confirmed the issue was fully resolved by 8:30 am UTC (2:00 pm IST), putting the total disruption at roughly three hours and 24 minutes. The outage affected claude.ai, the Claude API, Claude Code, and Claude Cowork, while Claude Console and Claude for Government remained unaffected throughout.
Downdetector numbers
Outage-tracking platform Downdetector recorded more than 1,300 user reports in the US alone during the disruption, while India logged over 180 reports of its own. Of the India reports, 55 percent of users flagged problems specifically with Claude Code, 21 percent reported issues using Claude's chat interface, and the remaining 14 percent said they were unable to use the Claude app.
Users take to social media
As the outage unfolded, affected users vented on X, with several posts referencing a '429 overloaded' server error that has reportedly grown more frequent in recent weeks. Anthropic had not issued a public statement in the early hours of the incident, directing users instead to its status page for updates, before formally acknowledging the outage once the cause had been identified.
Part of a pattern
Monday's outage was the latest in a string of service disruptions Anthropic has faced through August. Anthropic has previously attributed some of its outages to demand outpacing available compute capacity, driven by rising enterprise use of Claude Code alongside a surge in consumer sign-ups.
