Claude Code Login Outage Locks Developers Out
Anthropic's Claude Code CLI suffered an OAuth authentication outage on March 11, locking developers out mid-work while the Claude API remained operational.
5 min read
At 2:44 p.m. UTC on March 11, Anthropic's status page posted an incident titled "Elevated errors on Claude.ai (including login issues for Claude Code)." Three minutes later, a follow-up confirmed that login and logout actions inside Claude Code were broken. By 3:29 p.m. UTC, Downdetector was tracking over 1,400 concurrent user reports. Developers mid-session had no warning and no workaround - just a timeout message and a dead CLI.
TL;DR
- Claude Code CLI login broke around 2:44 p.m. UTC on March 11, with OAuth timeouts and 500 errors
- Over 10,000 users hit Downdetector to report problems
- The Claude API (direct key access) stayed operational all through
- Anthropic confirmed the issue but gave no fix timeline
- Community members patched the OAuth timeout threshold themselves within minutes
What Happened
The failure hit OAuth authentication - the login flow that connects Claude Code to Anthropic's servers through a browser-based authorization step. Users running /login in the CLI would see their browser open, click "Authorize," receive a confirmation message, and then watch the CLI hang until it timed out at 15 seconds.
The error was consistent across platforms:
OAuth error: timeout of 15000ms exceeded
Some users saw different failure modes. Authentication tokens that were already active expired without renewal. Others got 401 responses:
{
"type": "error",
"error": {
"type": "authentication_error",
"message": "Invalid authentication credentials"
}
}
Anthropic's status page confirmed elevated errors affecting Claude.ai and Claude Code login actions.
Source: techradar.com
The Claude.ai web interface was also affected, with users reporting blank screens and "Something went wrong" errors during login attempts. But the Claude API - the direct endpoint that developers access with API keys rather than OAuth - remained fully operational throughout.
That separation is the core architectural detail. Claude Code authenticates through OAuth, not through the API. When the OAuth infrastructure went down, every developer who depended on the CLI's browser-based login flow was locked out, even though the underlying model inference service was running fine.
What Developers Experienced
GitHub issue #33239 - Opened at 3:43 p.m. UTC by user nathan-brainkey, the bug report documented the failure on Claude Code v2.1.72 running on macOS. Within minutes, more than 15 users confirmed identical behavior. The flow was the same every time: run /login, browser opens, click authorize, browser says "you're all set," CLI shows the OAuth timeout.
Hacker News thread - A discussion thread surfaced reports from users on paid Pro plans experiencing not just login failures but degraded performance on sessions that were already authenticated. One user reported their VS Code extension running "extremely slow" while on a deadline. Another noted that the official status page had been showing green even as Downdetector was already registering a spike - a lag that eroded trust in the status page itself.
Downdetector tracked over 1,400 concurrent reports as the outage spread, peaking around 4:19 p.m. UTC.
Source: techradar.com
X/Twitter - Users coined "Clautage" to describe the event. One developer posted about being stuck during a major refactor, unable to resume work. The complaints tracked a consistent pattern: no advance warning, no graceful degradation, and no failover to API-key authentication inside Claude Code.
The Community Fix
The most notable detail wasn't Anthropic's response. It was the community's.
Within 11 minutes of the GitHub issue being opened, a user named doctorfarhan posted a workaround: manually editing Claude Code's compiled JavaScript to extend the hardcoded OAuth timeout from 15 seconds to 45 seconds.
// Before (in cli.js)
timeout: 15000
AbortSignal.timeout(15000)
// After (patched)
timeout: 45000
AbortSignal.timeout(45000)
On macOS, another user confirmed the patch worked after re-signing the modified binary with codesign -s -. The fix suggests that Anthropic's OAuth endpoint wasn't completely dead - it was responding, just too slowly for the client's hardcoded threshold. A longer timeout let the handshake complete.
That a 15-second hardcoded timeout in a compiled binary was the point of failure raises its own questions. Claude Code has become a core development tool for many teams. Its OAuth security model was tightened earlier this year exactly because of how sensitive the authentication flow is. A single non-configurable constant shouldn't be the difference between a working tool and a locked-out developer.
A Pattern Forming
Users encountered various error screens, from OAuth timeouts to generic "Something went wrong" messages.
Source: techradar.com
This isn't Claude Code's first outage in March. A larger incident on March 2 took down Claude.ai, mobile apps, and Claude Code for about 14 hours, with Anthropic citing "unprecedented demand." Before that, a February 3 outage was traced to an "internal routing error." In late January, users reported security vulnerabilities in the tool that raised separate concerns about its authentication architecture.
Three significant disruptions in six weeks isn't a fringe complaint. Anthropic's status page lists model safety and transparency, energy ratepayer protections, and infrastructure investments among its public policy priorities. Infrastructure reliability is conspicuously absent from that list.
Anthropic's official statement was limited:
"We're currently investigating issues with Claude Code and Claude.ai. Some users may be unable to log in, and others may experience slower than usual performance. The Claude API is not affected."
No root cause, no estimated time to resolution, and no post-mortem has been published.
What Developers Should Do Next
- Keep an API key configured as backup. Claude Code's OAuth login is a single point of failure. The API itself stayed up. Developers who had direct API key access were unaffected.
- Watch the status page, but verify with Downdetector. Multiple users reported that status.claude.com was still showing green while the outage was already spreading. Trust but verify.
- Pin your Claude Code version. The community workaround required editing a specific version's binary. Auto-updates during an outage can reset manual patches.
- Track GitHub issues in real time. The fastest diagnosis and fix came from GitHub issue #33239, not from Anthropic's official channels.
Sources:
