- Sources: Anthropic incident, OpenAI status history, discussion, second thread, third thread
- Summary: Anthropic's incident page runs from investigating at 13:26 UTC to impact ended at 16:16 UTC, names Mythos and Fable 5.1, Mythos and Fable 5, Opus 5, Opus 4.8 and Opus 4.6, and lists claude.ai, api.anthropic.com, Claude Code and Claude Cowork as affected. OpenAI's status history carries "Elevated errors across ChatGPT and Codex" resolved at 4:55 PM on 2026-09-03, with a note that some Codex remote control users may need to pair their mobile device again. That page carries no timezone label and no incident start, so the length of OpenAI's window and its relation to Anthropic's cannot be derived from the source, and what is stated here is that both ran on the same date. Neither vendor states a cause and neither references the other.
- Comments: HN commenters in the two largest threads report Grok failing the same day. status.x.ai declares no incidents and exposes no readable history, so that report is unverified here and no third vendor is claimed.
- Why it matters: Anthropic's incident covered every model it names plus the Claude API, Claude Code and Claude Cowork, so the usual failover target was a vendor logging its own errors across ChatGPT and Codex on the same date.
- Follow-up: Track whether either vendor publishes a postmortem naming a cause and an incident start time, and whether xAI publishes any incident record for this date.
send feedback on this story