Claude experiences outage, Anthropic deploys fix

Anthropic's public status dashboard reported "elevated error rates" across multiple Claude models on June 23, 2026. Per The Next Web, the status page shows Anthropic began investigating at 14:19 UTC and moved to "identified" by 14:25 UTC, with a fix being implemented. Affected services included Claude.ai, Claude Console, Claude API, Claude Code, and Claude Cowork, per The Next Web and Anthropic's status page. CNET and TechRadar reported over 8,000 user-submitted outage reports in the US via Downdetector at peak.

CybersecurityNews documents a separate outage the prior day, placing the start of investigation at 00:37 UTC on June 22 with a staged model-by-model recovery completing by 02:06 UTC - a roughly 90-minute disruption. That incident affected Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5, per CybersecurityNews. The June 23 incident follows this sequence by less than 14 hours, making the two events part of a broader pattern of instability in June.

CybersecurityNews describes a model-by-model recovery sequence for the June 22 event, with Opus 4.8 and Haiku 4.5 among the earlier models to recover. Per The Next Web, the June 23 incident was identified within six minutes of investigation opening - suggesting fast operational response - but it occurs alongside Anthropic having already suspended access to Claude Mythos 5 and Claude Fable 5 on June 13 for separate reasons.

Companies operating hosted model APIs commonly use staged rollouts for fixes to limit blast radius; the per-model recovery sequences described in coverage are consistent with that pattern. For production integrations, even short elevated-error windows can cause cascading failures in automation and CI/CD pipelines. The Next Web notes that Reddit communities report users building around provider instability rather than assuming it away, which signals a broader reliability expectation gap.

For practitioners: monitor the Claude status page (status.claude.com) for post-incident updates and model-specific advisories; implement client-side retries and exponential backoff if not already in place. Industry observers will watch whether incident frequency changes month-to-month and whether Anthropic publishes root-cause analyses for recurring disruptions, per The Next Web.