Anthropic reports elevated errors on three Claude models

Anthropic posted an incident report on September 22, 2026, flagging elevated errors across three AI models. At 00:57 UTC, it said requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 were failing more often. By 01:17 UTC, it had found the cause and was rolling out a fix.