What Changed

  • Scope: Incident expanded from Opus 4.6 alone to both Opus and Sonnet 4.6, per Anthropic’s Statuspage [2][1].
  • Status: Moved from “Investigating” to “Identified,” signaling Anthropic has isolated the issue but not resolved it yet [2][1].
  • Severity: Remains “Major,” with elevated error rates persisting across affected variants [1][2].

Cross-Source Inference

  • Broadened impact to Sonnet 4.6 plus continued major severity implies risk across both API and web app experiences where these variants are selectable (medium confidence), combining scope notes in the identified update [1] with the earlier Opus 4.6 investigation [2].
  • Shift to “Identified” without resolution suggests partial containment work is underway but production instability persists; no ETA is implied (medium confidence), based on Statuspage phase change [2→1] and unchanged severity labeling [1].

Implications and What to Watch

  • Now: Keep automatic failover to unaffected models/providers; apply conservative rate limits; surface user-facing degradation banners for Opus/Sonnet 4.6. Avoid traffic ramp-up until stability signals appear (high confidence) [1][2].
  • Safe resumption criteria: Statuspage moves to “Monitoring” or “Resolved” with a stated root cause or containment description, and no ongoing major impact (high confidence) [1].
  • Watch next: Statuspage entries labeled “contained,” “monitoring,” “resolved,” or “root cause”; any engineering follow-up clarifying variant-specific remediation steps (medium confidence) [1].
  • Correlation: No credible third-party telemetry indicating broader provider or shared-infra issues; treat as Anthropic-local until contrary evidence emerges (low confidence) [1][2].