Anthropic's Claude AI chatbot went down for more than 12,000 users on Tuesday morning, 29 September. Reports spiked at around 10 am ET, according to Downdetector.
Users attempting to log in to Claude were met with a message reading 'account temporarily unable to authenticate, please retry.' The error blocked access entirely for many, while others reported severely degraded performance across Claude's three main products: Claude Chat, the Claude application programming interface (API) used by developers to integrate the assistant into third-party applications, and Claude Code, the company's AI-powered coding tool.
Anthropic confirmed it was investigating the issue at 10:21 am ET. By 10:40 am ET, the company said it had applied a mitigation to reduce error rates, though it noted a full resolution was still in progress.
Claude Code Users Report Lost Sessions and Routines
The disruption hit developers using Claude Code hardest, who reported losing active sessions and automated routines mid-task. For engineers relying on Claude Code as a primary development tool, the authentication failure interrupted work that had no straightforward manual fallback.
Tuesday's disruption is not the first instance of its kind. Claude.ai logged 30 incidents between 12 August and 25 September this year, according to Deployflow's tracking. Of those, 18 occurred in August and 12 in September, which is before the 29 September outage added to the tally.
The Next Web reported that September alone had brought a series of outages marked by high error rates, describing the month's pattern as a recurring issue for Anthropic.
Claude's Security Incidents
The outage came amid broader scrutiny of Claude's operational security. Anthropic disclosed earlier in 2026 that four incidents of unauthorised access had occurred during cybersecurity evaluations. Claude models were mistakenly connected to the open internet due to a misconfiguration in the evaluation environment.
One of those incidents, disclosed in January 2026, involved Claude Opus 4.6. Anthropic published its account of those incidents on its own research pages, making it a primary disclosure rather than a third-party claim.
Anthropic had not issued a public root-cause statement as of the time of reporting.
