Claude Code Sign-In Outage, 13 Nov 2026: A Recovery Plan for Vibe Coders

What happened to Claude sign-in on 13 November 2026?

Anthropic's status page said it was investigating "an issue with users authenticating to Claude.ai and Claude Code" ^[1]. Business Today reported that the company acknowledged elevated authentication errors at 9:58 UTC, or 3:28 PM IST ^[2].

ChainCatcher reported that, after investigating, Anthropic had deployed a fix and was monitoring recovery ^[3]. The status page later marked the incident resolved at 11:58 UTC, for the issue affecting Claude.ai and Claude Cowork ^[1]. The resolved note does not name Claude Code, so do not assume Claude Code recovered at the same minute. Check your own sign-in before you resume work.

The timeline, as the sources give it, is short:

• 09:58 UTC: Anthropic acknowledged elevated authentication errors and began investigating ^[1][2].
• 11:05 UTC: A fix was deployed and recovery was being monitored ^[1][3].
• 11:58 UTC: The status page marked the issue resolved for Claude.ai and Claude Cowork ^[1].

Business Today also reported that Downdetector recorded 2,895 reports about Claude at around 3:45 PM on Friday. Those are user reports, not a count of affected users ^[2].

Which Claude products were affected, and which stayed up?

According to Business Today, the status page showed a major outage for Claude.ai. Claude Code and the Claude Console, which is served from platform.claude.com, were marked as partial outages. Claude Cowork reported degraded performance ^[2].

Why does a sign-in failure stop a vibe coder?

Building by prompting depends on one thing that an outage removes first: a live session. Claude Code reads your project, edits files and runs commands only while it can start a session. In our view, a sign-in error blocks new prompts, while the files on your disk stay as they were.

In our view, this matters more for vibe coders than for developers who write most of the code by hand. A developer can keep reading, planning and testing offline. A vibe coder's main way of making progress is the prompt, and the outage removes it.

What should you do when Claude Code will not sign you in?

Work through these steps in order.

1. Check the status page first. If there is an active incident, waiting is the right move. Avoid changing settings while an incident is open.
2. Do not delete credentials or reinstall the tool while the incident is open. Changing credentials during an incident makes it harder to tell the outage from your own changes.
3. Save your work where it stands. Commit or stash what you have, so the next session starts from a known state.
4. Write down the next prompt. Copy the prompt you were about to send and the last change the agent made. You can resume from the notes, not from memory.
5. Retry at intervals. Try once every few minutes, and log each attempt, so you can see when it starts working.

If a deadline falls during the outage, tell the people who depend on the work what has stopped and when you expect to resume. Tell them what stopped and give a time you can keep.

If the outage runs long and the work cannot wait, a second AI tool can handle small, contained edits. Keep those edits in a separate branch, so you can compare them with the main line once Claude Code is back.

How do you keep a project safe during an outage?

Set up these habits before the next outage. Three are enough to start.

• Commit at each working point. A commit after every feature or fix gives you a point to return to if an agent change goes wrong.
• Keep a short notes file. Record the goal, the current state, the last prompt that worked and the next step. Put it at the top of the repository.
• Keep secrets out of prompts. Store API keys in environment variables or a secret manager, and never paste them into a chat. An outage is a bad time to rotate a key you have leaked.

None of these steps needs the service to be up. They work offline and cost a few minutes a day.

What should you check after the service recovers?

When sign-in returns, do not go straight back to large changes. Start with a small test.

• Sign in fresh, then send one small prompt on a throwaway branch, such as a request to explain one file.
• Read the diff of the last session before the outage. Check whether any change from the interrupted session is complete.
• Run your tests and your build before you merge anything that was written during the incident.

Keep the notes file current until you have worked normally for a while.

What is still broken?

The sign-in incident was not the only problem that week. Business Today reported that a Microsoft 365 connector problem was separate and still ongoing. "Anthropic said a Microsoft 365 service incident was affecting its mail, calendar and search tools" ^[2]. Most requests to those tools continued to fail, and Claude for Outlook was also affected.

If your workflow depends on the connector, do not assume it is fixed because the sign-in incident is resolved. Test the connector separately.

How do you set up a fallback before the next outage?

A fallback is a plan you can run without thinking during an incident. It does not need to be elaborate. Four parts are enough.

• A second tool for small edits. Choose one other AI assistant that can read a file and propose a change. Use it only for contained fixes, such as a copy change or a single function.
• A local test command. Keep one command that runs your tests and build. If the agent is down, you can still tell whether your code works.
• A written prompt library. Save the prompts that worked for your project: the setup, the schema, the styling rules. A new session can then start from the same instructions.
• A rule for what waits. Decide in advance which tasks wait for the tool to return, such as large refactors, and which can continue, such as writing notes or fixing a typo.

Write these down once, in the same notes file you keep for the project. The first time you need them should not be during an outage.

Is a single AI coding tool a risk for a vibe coder?

It can be. In our view, the sign-in incident shows the pattern: the tool stops, and the work stops with it. Business Today's report and the status page both describe a problem that was limited in time, and the status page recorded the issue as resolved the same day ^[1]. That does not make every outage minor, and it does not tell you how often the next one will come.

The practical question is how much of your project lives only inside the tool. If the answer is "all of it", the risk is larger than the outage itself. Keeping the code in your own repository, with your own history, reduces that dependence. The tool then becomes a faster way to write code, not the only copy of the project.

This is a judgment about working practice, not a finding from the sources. Adjust it to the size and stakes of your project.

What should go on your checklist for the next outage?

Keep this list in your notes file, so you can run it without looking for it.

• Check the status page before you change anything.
• Save your work: commit or stash, and write down the next prompt.
• Retry at intervals and log each attempt.
• When sign-in returns, send one small test prompt on a throwaway branch.
• Read the last session's diff, then run the tests before you merge.

The list is short on purpose. A longer one is harder to follow when the tool you rely on is the thing that has stopped.

What do readers ask about this outage?

What time did Anthropic acknowledge the problem?

Business Today reported 9:58 UTC, or 3:28 PM IST ^[2]. The status page recorded the same incident as investigating at that time ^[1].

When was the incident marked resolved?

The status page marked the issue affecting Claude.ai and Claude Cowork as resolved at 11:58 UTC ^[1]. The note does not name Claude Code.

What should I do first if I cannot sign in?

Check the status page, then wait, and save your work locally. Do not reinstall the tool or change credentials during an active incident.

Does a resolved incident mean everything works?

Not always. Test the tools you use, because a resolved note may cover only some of the products that were affected.