Anthropic began investigating an authentication incident on November 13, 2026 that left claude.ai in a major outage and Claude Code in a partial outage, according to the company's status page. The first public note landed at 09:58 UTC, saying users could not authenticate to claude.ai and Claude Code. The same board still showed api.anthropic.com operational, which matters for teams that route production traffic through the API rather than the web UI.
If you depend on Claude for daily coding, this is the second reliability shock in forty-eight hours for Anthropic customers. On November 12, a Microsoft 365 connector incident kept mail, calendar, and search tools failing upstream. Today's break hits login itself, which stops IDE sessions cold even when models are healthy.
TL;DR #
| Surface | Reported status (Nov 13, morning UTC) |
|---|---|
| claude.ai | Major outage |
| Claude Console (platform.claude.com) | Partial outage |
| Claude API | Operational |
| Claude Code | Partial outage |
| Claude Cowork | Degraded performance |
| Trigger | Authentication failures |
| First post | Nov 13, 2026, 09:58 UTC |
Official timeline from status.claude.com #
Anthropic's incident history for November 13, 2026 lists an unresolved item: elevated errors on claude.ai, Claude Code, and platform.claude.com. The live banner at publish time stated:
We are investigating an issue with users authenticating to Claude.ai and Claude Code. We will provide an update as soon as possible. Nov 13, 2026 - 09:58 UTC
That language is narrower than a full platform blackout. It points to identity and session establishment, not model inference. Yet the status summary still marked claude.ai as major outage, which usually means most web users cannot complete normal work.
The API green check is the critical split:
- Batch jobs, eval harnesses, and server agents that hold long-lived API keys may keep running.
- Claude Code, Cowork, and browser chat depend on OAuth and web session flows that fail when authentication breaks.
None of the posts reviewed here gave a root cause, a percentage of affected users, or a recovery ETA.
Separate incident: Microsoft 365 connector still open #
Do not merge today's auth problem with yesterday's connector outage. Anthropic's status feed still carries November 12 updates saying mail, calendar, and search tools on the Microsoft 365 connector fail for most requests, with Claude for Outlook affected. Anthropic attributes that thread to a Microsoft 365 service incident and points readers to status.cloud.microsoft.
Our standalone write-up on Exchange Online EX1550543 explains how Microsoft health-check failures cascaded into Copilot and mail pain on the same day. Agents that chain Claude Meetings or Outlook add-ins with Claude Code now face two independent failure modes: Microsoft cloud health and Anthropic authentication.
What developers are saying on Hacker News #
A Hacker News thread opened as the outage spread. The discussion is thin compared with major launch threads, but two comments are still useful as unverified user reports:
- User cheptsovlinked to the official status page as confirmation.
- User icodzreported a few minutes of failure showing a** Cloudflare block** , then recovery after changing Wi-Fi networks.
Treat network-path anecdotes as hints for support triage, not proof of root cause. They do suggest some failures may sit in front of Anthropic's own services.
Why authentication outages hurt agent workflows hardest #
Modern Claude use is not "open the website and type." Teams wire Claude Code into CI, Cowork into document review, and console projects into org-wide keys. Authentication sits on the critical path for:
- IDE extension refresh — Claude Code must renew tokens to call tools and subagents. A stuck login looks like "the model broke" even whenapi.anthropic.com is green.
- Org seat provisioning — New hires cannot onboard during auth outages, which blocks the exact week many enterprises expand agent pilots afterAnthropic's IPO roadshow .
- Cross-product demos — Cowork plus Code plus Meetings storylines fail when only the API survives; degraded Cowork on the status board matches that pain.
Reliability planning should assume UI and API diverge. Our guides on token-efficient agent reading and Claude Code observability after 2.1.323 assume you can still run sessions. Today is a reminder to log which surface failed when post-morteming agent throughput.
Login failure metaphor for a Claude authentication outage on claude.ai and Claude Code
Illustration: when authentication fails, web and IDE paths drop even if API gateways stay up.
How this compares with earlier November Claude incidents #
Anthropic's status history for the month is already crowded. Treat each line as a separate blast radius:
| Date | Incident shape | Primary user pain |
|---|---|---|
| Nov 8 | Degraded Fable 5.1 | Model errors across web, API, Code, Cowork for ~14 minutes |
| Nov 10 | Overload on Fable 5.2 / Mythos 5.2 | Short API errors on newest SKUs |
| Nov 12–13 | Microsoft 365 connector | Tool failures on mail and calendar, not login |
| Nov 13 | Authentication | Cannot sign in to web and IDE surfaces |
The November 13 event is worse for interactive builders because it blocks session refresh. You cannot open a new Claude Code workspace, reauthorize Cowork, or invite a teammate while authentication is red, even if yesterday's prompt templates still live on disk.
Teams that standardized on Claude Code plus Meetings after November office-suite launches should document which SKUs each workflow needs. A meeting summary agent might still run on API keys while facilitators cannot log into claude.ai to edit the project instructions.
Enterprise procurement questions to ask vendors #
When your renewal hits during outage season, ask Anthropic or your reseller:
- Separate SLAs for claude.ai, Claude Code, and API — are they the same contract clause?
- Status communication — will authentication incidents get distinct postmortems from connector incidents?
- Failover — does your org already holdBedrock-hosted Claude orVertex capacity for critical jobs?
- Audit logs — can security teams exportlogin failure rates without exposing message content?
None of those questions change today's incident. They prevent tomorrow's incident from becoming a revenue halt.
Practical fallbacks for builders #
If you ship on the Claude API
- Keep health checks on api.anthropic.com , not only claude.ai.
- Route critical automation through server-side keys with retry and circuit breakers, the same pattern we recommend inBedrock cost baselines .
- human-in-the-loop merges that depend on Claude Code until auth recovers. Partial outages are worse than full ones because some teammates still work while others cannot sync.
If you live in Claude Code
- Retry after status.claude.com shows recovery; do not reinstall extensions first.
- Switch networks if you see edge blocks similar to the HN report above.
- Fall back to API-direct harnesses (your own scripts, OpenRouter, or Bedrock-hosted Claude) only when your contract allows the same model ID.
If you run multi-vendor agents
- Parallel Codex on Bedrock orGemini CLI capacity is not free, but acold standby model beats a full stop during a login incident.
Context: Anthropic reliability in November 2026 #
This month already includes:
- November 8 — A shortClaude Fable 5.1 degradation across claude.ai, API, Code, and Cowork.
- November 10 — Brief overloaded errors onFable 5.2 andMythos 5.2 .
- November 12–13 — Microsoft 365 connector failures tied to Exchange health checks.
The pattern is not unique to Anthropic. Cloud AI stacks bundle identity providers, connector marketplaces, and model serving. Any one layer can fail independently. Enterprises negotiating Anthropic Meetings and Sheets should ask vendors for incident IDs per dependency, not a single "all green" dashboard.
Security note during outages #
Outages attract fake status pages and credential phishing. Verify links against status.claude.com only. Our coverage of fake OpenAI and Anthropic event invites shows how quickly attackers copy outage panic. Do not paste API keys into unofficial "recovery" forms.
What this does not change #
Nothing in the November 13 status posts alters model pricing, Fable 5.2 list cuts, or export-control politics around Nvidia and China chips. Leadership churn at Nvidia (Huang's resignation) does not touch Anthropic's auth stack directly.
When service returns, resume normal prompt regression tests and agent eval baselines. One auth blip should not rewrite your model choice, but it should update your runbook.
Post-incident checklist for agent leads #
After status.claude.com clears authentication:
- Rotate nothing blindly — auth outages are not token leaks. FollowApiiro-style guidance on settings.local.json only when secrets were exposed, not when login failed.
- Replay failed CI jobs — note which jobs used OAuth versus API keys; fix the OAuth path first.
- Update runbooks — link this post and the official incident ID in your internal wiki.
- Communicate to legal — if you d customer-facing agents, log downtime for SLA credits separately from model quality regressions.
Monitoring signals that help next time #
Set external synthetics that hit three URLs:
https://claude.ai/loginor your org SSO entryhttps://api.anthropic.com/v1/messageswith a trivial prompt- Your Claude Code health check script
Divergence between the second and first results is the early warning for exactly this incident class. Pair with SparseRead-style token budgeting so retries during partial outages do not burn quota.
Status timelines and component names are accurate as of November 13, 2026 UTC; check status.claude.com for recovery.