Live  /  Google Cloud  /  Incident

Google Cloud: Vertex AI Gemini API customers experienced increased error rates when accessing the global endpoint.

Google Cloud · cloud.google.comResolved2026-03-09 05:25 UTC
Public — what happened

Summary

Google Cloud said that on 27 February 2026, starting at 04:37 US/Pacific, customers calling Vertex AI Gemini models saw elevated error rates through the global endpoint and in US regions, with knock-on problems for Dialogflow CX, Agent Assist and Google Cloud Support. The company reported that pay-as-you-go users mostly got 429 errors while Provisioned Throughput users got 503 errors. Google Cloud attributed the disruption to a configuration change to a safety filtering service used by all Gemini models, and said rolling that change back and adding capacity restored full service by 06:35 US/Pacific, roughly 1 hour and 58 minutes after it began. It said it plans to add mandatory validation checkpoints to rollouts and to strengthen alerting on critical dependencies.

Timeline

Source of record: provider status page. Summary written by SquidGrid from the provider’s published updates; times in UTC.

Independent path health

The status page is the symptom. SquidGrid measures the paths that reach Google Cloud against their own baselines — whether the routes carrying your traffic degraded, at which handoff, and who owns that stretch.

🔒 Sign in to see
Customers get the live Provider Assessment Report for Google Cloud: paths off baseline right now, where on the route conditions change, and its full incident track record.

Included with any paid plan.

Open the Google Cloud report →

More from Google Cloud

The full archive is on the Google Cloud status & history page.

Related: Google Cloud status & history. SquidGrid is not affiliated with Google Cloud.