ChatGPT, Claude, Grok and Codex all went down inside the same two-hour window.

Starting around 7:57 AM PT on September 3, all four reported outages simultaneously, with Downdetector logging 35,000+ US reports for ChatGPT, 1,400 for Claude and 1,200 for Grok before recovery by 12:38 PM PT. Cloudflare denied any significant disruption and xAI traced its own failure to a Memphis data center, so the shared-infrastructure theory is unconfirmed and the simultaneity may be coincidence. Either way it's the first day the entire agentic tooling layer was unavailable at once, and any product whose core loop is an LLM call has a correlated-failure risk that multi-vendor fallback does not obviously fix. The Register