Claude is down, status page says all systems operational (status.anthropic.com)

Anthropic says the recent “Claude” quality problems were caused by two separate software bugs, not throttling or intentional degradation, and that fixes and mitigations have been deployed. The company traced one issue to a bug that affected a small percentage of Claude Sonnet 4 requests between Aug. 5 and Sep. 4 (with impact rising from Aug. 29–Sep. 4), and a second bug that impacted some Claude Haiku 3.5 and Sonnet 4 requests from Aug. 26–Sep. 5; a related impact on Opus 3 was also identified. Status pages now report no ongoing problems, and a technical post‑mortem will be published next week. For the AI/ML community this underscores two points: (1) even mature model fleets can exhibit subtle regressions that only appear in production traffic patterns, and (2) community reports and robust monitoring are vital for catching quality degradations quickly. Practically, teams using Claude during the affected windows should validate any critical outputs created then; developers should watch the upcoming post‑mortem for root‑cause details and mitigation patterns that could inform observability, regression testing, and deployment safeguards in their own model pipelines.