Claude Code is locking people out for hours
TL;DR Highlight
Claude Code is experiencing repeated service stability issues such as OAuth timeouts, query slowdowns, and malfunctioning background agents. Concerns are growing that this is not simply a bug, but a structural problem related to Anthropic's compute capacity limits.
Who Should Read
Developers actively using Claude Code (including Max subscriptions) in their workflows, who are experiencing significant productivity impacts due to service stability issues or are considering alternatives.
Core Mechanics
- When logging into Claude Code in a Windows + WSL environment, a bug repeatedly occurs where the 'OAuth error: timeout of 15000ms exceeded' error appears, making the app unusable. This is as of version 2.1.92, and the Google account login flow is not completed.
- This login failure coincides with a large-scale service outage that occurred on April 6, 2026, and is not an isolated bug. Most Claude Code users were unable to log in. The status of the outage could be checked at status.claude.com.
- Multiple reports indicate that Claude Max subscribers are exceeding their limits after just one query. Community analysis suggests that Max users are grouped into a shared compute pool, and the surge in demand following issues with OpenAI has exceeded the pool's upper limit.
- There are reports that query response speeds have slowed down by approximately 3x compared to before. Both Sonnet and Opus are reportedly slow in the Claude Code on Max environment.
- There are complaints that the background agent functionality is essentially broken. When a task is assigned to the Sonnet background agent, it fails due to permission errors, and the expensive Opus model intercepts and processes the task in the foreground. This impacts both context management and token costs.
- One community comment interpreted this phenomenon as 'distillation' (a technique for transferring the capabilities of a large model to a smaller model). It suggests that Anthropic is degrading quality by replacing Claude Code with a lightweight model to reduce compute pool load, and subjective experiences of increasingly frequent unnecessary interruptions like 'Continue?' or 'Run this command to see this information' have increased in recent times after being absent for over a year.
- A practical tip shared is that if a locked state has not yet occurred, pre-preparing a token with the 'claude setup-token' command allows bypassing login issues by using 'CLAUDE_CODE_OAUTH_TOKEN=sk... claude' in the future.
Evidence
- "After this issue became a hot topic on Hacker News, a comment pointed out that the original title '[BUG] Claude Code login fails with OAuth timeout on Windows' was much more specific, but an HN editor changed it to 'locking people out for hours,' which was criticized as an exaggeration. Other comments countered that the exaggeration was not justified considering the issue arose during a large-scale outage."
How to Apply
- If you can still log into Claude Code, run the 'claude setup-token' command now to save the OAuth token in advance. This will allow you to bypass access issues by using 'CLAUDE_CODE_OAUTH_TOKEN=sk... claude' in the future if a login failure occurs.
- If you are operating Claude Code on Max on a team basis, include status.claude.com in your monitoring pipeline and prepare a fallback flow to directly call the Anthropic API when an outage occurs to minimize service disruption.
- If background agents are failing and Opus is performing the task in the foreground, wasting context and tokens, recognize that agent permission processing is currently unstable and explicitly specify the model and execution method for important long-term tasks, or manually separate and execute tasks for safety.
- If you are using Claude Code in business-critical workflows, it is recommended to use a stable release channel or pin a specific version instead of automatically updating to the latest version of Anthropic, and to develop a habit of monitoring for performance degradation due to external factors such as compute capacity issues.
Code Example
# 1. Generate token when login is possible
claude setup-token
# 2. Bypass execution with token when login failure occurs
CLAUDE_CODE_OAUTH_TOKEN=sk-ant-xxxxx claudeTerminology
Related Papers
Formal Verification Gates for AI Coding Loops
AI가 생성한 코드에서 보안 불변식(invariant)을 지키게 하려면 프롬프트 지시보다 타입 시스템 같은 구조적 제약이 훨씬 효과적이라는 주장과 구현 방법을 소개한다.
Learnings from 100K lines of Rust with AI (2025)
Azure RSL(분산 합의 라이브러리)을 Rust로 재구현하면서 AI 코딩 에이전트를 활용해 4주 만에 100K 라인을 작성한 경험담으로, Code Contracts와 Spec-Driven Development를 AI와 조합하는 실전 워크플로우를 공유한다.
A Methodology for Selecting and Composing Runtime Architecture Patterns for Production LLM Agents
LLM agent가 왜 터지는지 이름 붙이고, 어떤 아키텍처 패턴을 언제 써야 하는지 5단계로 정리한 실전 가이드
Show HN: Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks
작은 로컬 LLM(8B)에 guardrails(구조적 안전망)를 씌워 멀티스텝 에이전트 작업 성공률을 53%에서 99%까지 올린 Python 프레임워크 Forge 공개. 모델 자체는 건드리지 않고 실행 환경을 강화하는 접근법이라 주목받고 있음.
Mini Shai-Hulud Strikes Again: 314 npm Packages Compromised
2026년 5월 19일, npm 계정 하나가 탈취되어 22분 만에 637개 악성 버전이 배포됐고, echarts-for-react·size-sensor 등 월 수백만 다운로드 패키지들이 감염되어 AWS 자격증명·SSH 키·AI 코딩 에이전트까지 탈취하는 정교한 공급망 공격이 발생했다.
Code as Agent Harness
LLM 에이전트에서 코드를 단순 출력물이 아닌 추론·행동·환경 모델링의 실행 인프라로 재정의한 102페이지짜리 서베이