Claude Outage: Claude.ai, API, Code and Cowork Report Major Errors

Anthropic is investigating a critical incident affecting Claude.ai, the Claude API, Claude Code and Claude Cowork. At 19:49 UTC on July 29, 2026, Anthropic's status page marked each of those services as a major outage and reported elevated errors across all models.

This is a live incident report. The status can change quickly; check Anthropic's official status page for the newest update before changing production workflows.

What is affected?

Anthropic's incident record names four affected components:

• Claude.ai — major outage
• Claude API (api.anthropic.com) — major outage
• Claude Code — major outage
• Claude Cowork — major outage

Claude Console (platform.claude.com) and Claude for Government were still shown as operational on the status page at the time of publication. That does not mean API requests, Claude Code sessions or individual models are healthy: Anthropic's incident says it is investigating elevated errors across all models.

Current official status

The incident page did not publish a root cause, an estimated recovery time or a workaround at the time of writing. Any claim beyond that would be speculation.

What developers should do now

1. Pause non-essential Claude-dependent jobs. Scheduled content generation, background coding tasks and bulk processing can compound retries and make incident diagnosis harder.
2. Use bounded retries. Do not let workers retry indefinitely. Apply exponential backoff, a retry cap and an alert once error rates cross the service-level threshold.
3. Protect queued work. Preserve idempotency keys and job state so a recovered request cannot duplicate a database write, email, deployment or payment action.
4. Fail over only if you have tested it. A provider fallback can keep a service available, but model behaviour, tool-call formats, safety settings and token limits differ. Route low-risk work first; do not silently send sensitive data to an unapproved provider.
5. Tell users what is happening. If your product exposes Claude-backed functionality, show a short incident notice rather than pretending a request is still processing.

Guidance for Claude Code users

Claude Code users may see failed requests, stalled turns or tool loops that cannot obtain a model response. Save local changes, avoid repeating destructive commands, and keep your shell session separate from an interrupted agent loop. If your repository uses an agent for migrations, releases or production operations, keep human approval enabled until the service is stable.

For teams running automated coding workflows, the safe temporary mode is read-only: inspect code, collect logs and prepare a change list, but do not auto-merge, deploy or modify infrastructure based on a partial or failed run.

Guidance for API teams

If your app calls api.anthropic.com, distinguish provider errors from your own validation failures in monitoring. Track status code, request ID, model, latency, retry count and final user outcome. A circuit breaker is useful when a dependency is known to be failing: it stops a burst of requests from becoming a retry storm and gives a clean fallback response.

Do not assume a successful HTTP connection means a completed job. Confirm that your application stored the expected result and that downstream actions were not triggered twice after a retry.

Is this a global Claude outage?

Anthropic classified Claude.ai, its API, Claude Code and Claude Cowork as major outages in the incident. The status page says the company is investigating elevated errors across all models. Console and Claude for Government remained operational in the component view at publication time, so the most accurate description is a critical multi-product Claude incident, rather than a claim that every Anthropic service is unavailable.

When will Claude be back?

Anthropic had not provided an estimated restoration time in the public incident update available at publication. Follow the official incident page for updates. This article will be revised when Anthropic identifies the issue, moves it to monitoring or declares it resolved.

Related guides

• MCP 2026: The Complete Developer’s Guide
• OpenCode: the open-source AI coding agent
• MiniCPM5-1B local AI-agent guide

Source

• Anthropic system status, incident “Elevated errors across all models,” checked July 29, 2026 at publication.