An error spike affecting multiple Anthropic AI models was identified early on September 22, 2026, prompting an immediate investigation by the company. Monitoring systems recorded a rise in failure rates for requests to Claude Mythos 5.1, Claude Fable 5.1, Claude Opus 5, and related services between 5:50pm Pacific Time and 7:10pm Pacific Time, corresponding to 00:50 to 02:10 UTC. Initial alerts indicated degraded performance across the Claude.ai platform, the public API at api.anthropic.com, and integrated tools including Claude Code and Claude Cowork.

By approximately 01:17 UTC, engineers confirmed they had isolated the root cause to a technical anomaly within the Mythos 5.1 and Fable 5.1 inference pipelines. Subsequent updates posted at 01:35 UTC noted a return to baseline success rates for Fable 5, Fable 5.1, Mythos 5, and Mythos 5.1, though residual issues persisted for Opus 5. A final status report at 02:11 UTC verified full restoration across all affected endpoints, with Anthropic stating that normal request handling had resumed.

The incident underscored the operational dependencies of downstream developers on stable model APIs, with the company emphasizing its commitment to transparency through real-time status updates. No external factors or third-party service disruptions were cited in the public timeline.