Elevated errors on claude.ai, Claude Code, and platform.claude.com

An outage affecting Anthropic’s Claude AI services, coinciding with a major Microsoft 365 failure, exposed how dependent many teams have already become on cloud LLMs for coding, ops, and everyday office work. Commenters trade outage reports from around the world, criticize fragile authentication flows (like email-only “magic links”) and single-vendor lock-in, and debate whether organizations should treat AI tools as convenience boosts or as critical infrastructure requiring redundancy and fallbacks. The incident also fuels broader worries about reliability, technical debt, and what it means for a soon-to-IPO AI provider to position itself as core productivity infrastructure while still struggling with frequent service disruptions.

Scope of the outage

  • Users across multiple regions (Europe, Middle East, etc.) reported Claude web/app issues, forced logouts, and failed logins.
  • Status page framed it as “elevated errors,” mainly around authentication and connectors (MCP/“Connectors” integrations like Google Docs).
  • Some could keep existing sessions alive or reconnect intermittently; others were blocked for hours.
  • There were reports of a second incident after an initial “resolved” notice; some features (e.g., connectors) were still degraded.

Authentication, login design, and Microsoft dependency

  • Many complaints about magic-link-only logins: inconvenient on mobile, brittle when the email provider (e.g., Microsoft 365) is down, no password+TOTP fallback.
  • Some users struggle to change email or are forced toward Google SSO that doesn’t match their domain setup.
  • Multiple people hit OAuth errors and suspected dependency on Microsoft Entra/M365; some confirmed they couldn’t log in because their corporate email (on M365) was down.
  • Debate on passkeys and SSO: some say passkeys exist but are rarely requested; SSO also broke during the incident.

Relation to broader Microsoft outages

  • Many noticed timing overlap with a major Microsoft 365 / Outlook outage and speculated shared infrastructure or indirect coupling.
  • It’s unclear from the discussion whether Claude runs directly on Azure or only rents some hardware; some think partial Azure usage could cause capacity or auth issues.
  • Some argue this shows the risk of deep reliance on a small set of cloud/identity providers.

Impact on work and AI dependency

  • Strong theme: developers and knowledge workers heavily reliant on Claude (and LLMs generally) felt “snow day”–like disruption.
  • Some openly say they now feel “pathetic” or uncomfortable working without AI, especially for code review, IaC, and boilerplate.
  • Others push back, arguing professionals should remain able to function without LLMs and that over-reliance is dangerous and immature.

Redundancy, reliability, and management

  • Repeated calls for:
    • Multi-provider failover (Claude, OpenAI, DeepSeek, local models, etc.).
    • Internal policies allowing more than one LLM vendor.
    • Self-hosted/local models as a resilience layer.
  • Some see Anthropic as having more outages than other LLM providers; others cite recent uptime stats as “solid” aside from this.
  • Outage timing raised concerns about Anthropic’s readiness to be a trillion‑dollar, IPO-bound backbone for enterprise workflows.

Broader reflections on AI in software development

  • Split views:
    • Enthusiasts claim 10–30x productivity gains, heavy agentic workflows, and AI as essential tooling.
    • Skeptics warn of “vibe-coded” spaghetti, technical debt, and codebases only AI can navigate.
  • Several compare this dependence to historical tech shifts (calculators, CAD), but note that SDLC safeguards and redundancy haven’t caught up.