What "Authentication service was unavailable" means
Read the message literally, because it is accurate. Two sentences, and they are doing different jobs.
Authentication service was unavailable. The system that checks who you are could not be reached. That is a statement about Anthropic's infrastructure, not about your password, your email address, or your subscription.
You can try again. This is the part that causes the damage. It is true in the narrow sense that the button works, and misleading in every other sense, because trying again sends the same request to the same unavailable service.
What makes it genuinely confusing is the behaviour around the message. You get signed out mid-session without warning. You enter your credentials, they appear to be accepted, and you are signed out again within seconds. Do that three or four times and the natural conclusion is that something has gone wrong with your account.
It almost never has.
Why "you can try again" does not help
Authentication and the rest of the product are separate systems. When the auth service is degraded, your session token cannot be validated, so the application does the only safe thing available to it and signs you out.
Logging in again sends a fresh request to the same unavailable service. Sometimes it succeeds, because the failure is often partial rather than total, which is why you occasionally get in for a few seconds before being dropped again. That intermittency is the most misleading part of the whole experience: partial success looks like a problem with your specific attempt rather than a problem with the service.
Clearing cookies, resetting your password, and reinstalling the desktop app all do nothing for this, and the password reset in particular is worth avoiding. A reset flow also depends on the auth service, so a reset started during an incident can fail halfway and leave you worse off than when you started.
How to tell whether the error is you or them
Check the status page first. It takes ten seconds and it saves you from troubleshooting a problem you cannot fix.
Two things to look for. The component list at the top tells you the current state of claude.ai, the Console, the API, Claude Code and Claude Cowork. Below that, the Past Incidents section shows anything reported in the last few days with timestamps in UTC.
If an authentication incident is open, stop troubleshooting. There is nothing on your side to fix and the only useful action is waiting.
One caveat worth knowing: the status page is updated by people, not automatically, so during the first few minutes of an incident it can still show everything as operational while the product is visibly broken. If the error is affecting you and the status page looks clean, give it five minutes and check again.
What to do while you wait
Nothing clever. The incident on 16 August 2026 was investigated at 21:58 UTC, escalated to degraded performance across every surface four minutes later, had a fix deployed by 22:22, and was resolved at 22:34. Thirty-six minutes end to end.
That is fairly typical for an auth incident. They are short, because authentication is critical infrastructure and it gets attention immediately.
If you want to know the moment it clears rather than refreshing the page, the status page has a subscribe option that sends email, SMS or Slack notifications when an incident is created, updated or resolved. It also publishes an RSS and Atom feed, which is the lower-friction option if you already read feeds.
If it persists after the incident closes
Once the status page shows the incident resolved and you are still locked out, then it is worth touching your own setup. In order:
Sign out fully rather than closing the tab, then sign in again, so you are issued a new token rather than retrying a stale one. If you use the desktop app or Claude Code, restart it, because both can hold a cached token that survived the outage. If you are on a corporate network or a VPN, try without it, since some auth flows are sensitive to routing. And if you are still stuck, Anthropic's support site is the right destination rather than another password reset.
The short version
Authentication service was unavailable. You can try again. The first sentence is the diagnosis and the second is a trap.
Anything that repeatedly asks for your password trains you to believe your password is the problem, which is why people reset credentials that were working fine. The error is telling you plainly that a service was unavailable. The status page confirms it in ten seconds, and that is the only action worth taking before you wait.