It wasn’t a big bang, but you could feel it: on August 4th, around 21:59 UTC, Anthropic flagged an issue on its status page. Some users couldn’t get in — logging in via OAuth to Claude.ai and Claude Code was failing. And those already inside sometimes saw errors on requests to the models.
What exactly happened
Two things were affected. First, sign-in: the OAuth flow that Claude.ai and Claude Code authenticate through threw errors for a portion of users. Second, the models themselves — Anthropic reported elevated error rates on requests to Mythos 5, Fable 5, Opus 5, and Sonnet 5. So across the current lineup, not just a single model.
The issue was identified and resolved, and success rates recovered. No data loss, no multi-day outage — more one of those disruptions you experience mid-work as a brief stutter and then forget.
My take
On its own, a hiccup like this is no big deal. It’s interesting as a reminder. If you’ve wired Claude — and especially Claude Code — firmly into your workday, every stutter like this shows how much already rides on it. Two things I’m taking away: first, it’s worth knowing Anthropic’s status page before you go hunting for the bug in your own setup. And second, it’s no accident that the login path becomes the weak spot — authentication is the bottleneck everyone has to pass through. If that doesn’t work, the best model in the world does you no good. Take a breath, check the status page, carry on. This time it was over quickly.
Sources: