Claude went down again on June 23, the third significant Anthropic outage in three weeks and the second of the day. An “elevated error rate across multiple models” incident opened at 14:19 UTC and wasn’t marked resolved until 17:44 UTC, hitting Claude.ai, the API, Claude Code, and Claude Cowork. Claude for Government stayed up. Earlier, starting at 06:33 UTC, a separate Opus 4.8 incident had already burned through an hour and fifteen minutes.
Downdetector logged 7,119 US reports at the 10:13 a.m. ET peak, per TechRadar; TechTimes put the broader total above 8,000. Anthropic moved the public incident from “investigating” to “identified” within six minutes, which The Next Web read off the status timeline as unusually fast triage. That’s the operational tell: the company knows where to look now.
The June ledger is the real story. June 13 saw Anthropic suspend access entirely to Claude Mythos 5 and Claude Fable 5, which The Next Web flagged as “a stronger signal than the company’s usual investigating language.” On June 16, between 17:23 and 18:00 UTC, Sonnet and Opus models ran at roughly a 10% error rate, with Opus 4.8 alone continuing to misfire until 19:20. June 22 brought a 1-hour-50-minute Opus 4.8 outage and a five-minute multi-model blip that evening. Then June 23, twice.
The numbers Anthropic itself publishes are the load-bearing detail. Trailing 90-day uptime sits at 99.12% for Claude.ai, 99.28% for Claude Code, and 99.41% for the API. Enterprise software contracts typically demand 99.9%, which permits roughly two hours of downtime per quarter; Anthropic’s figures imply 19 to 23 hours per service. Thoughtworks, writing after the June 2 outage, called hardcoding a single provider endpoint “a single point of failure that’s a very real threat to business continuity.”
Anthropic told Fortune that demand had grown faster than infrastructure could support at peak hours, and attributed at least one recent failure to a sub-agent bug in which delegated tasks multiplied exponentially rather than terminating, what TechTimes described as “a documented failure mode when an orchestration layer lacks cycle detection.” Annualized run rate moved from $9 billion at the end of 2025 to $47 billion by May 2026. The IPO filing values the company at $965 billion.
That’s the structural bind. The reliability disclosures sit inside the S-1 universe now, where they get parsed by analysts who price SLA risk for a living, not by users who can refresh and move on. Anthropic hasn’t published a root-cause analysis for any of its June 2026 outages.
Sources
- https://status.claude.com/
- https://www.techradar.com/news/live/claude-down-june-23-2026
- https://thenextweb.com/news/claude-outage-anthropic
- https://www.techtimes.com/articles/318925/20260623/claude-outage-tops-8000-reports-agentic-pipeline-failures-mount-before-anthropic-ipo.htm
- https://www.techedt.com/claude-outage-affects-users-as-anthropic-investigates-service-disruption