Incident Report

Summary

On Friday, 27 February 2026 at 04:37 US/Pacific, customers using Vertex AI Gemini API models experienced increased error rates. Impacted services included Google Cloud Support, Agent Assist, Vertex Gemini API and Dialogflow CX in US regions and the global endpoint. The issue persisted for a duration of 1 hour and 58 minutes.

This is not the level of quality and reliability we strive to offer you, and we have taken immediate steps to improve the platform’s performance and availability.

Root Cause

This incident was caused by a configuration change to a safety filtering service that supports all Gemini models. For some specific requests, this created code paths that eventually led to service disruptions and capacity loss for the safety filtering service. Consequently, customers encountered overload (429 and 503) errors for their queries, with some users reporting elevated error rates for specific models in US regions.

Remediation and Prevention

Google engineers were alerted to the issue via our automated monitoring system on Friday, 27 February 2026 04:54 US/Pacific and immediately started an investigation.

Engineers identified the faulty configuration change and initiated a rollback to restore the previous stable configuration. Engineers also added more capacity to the service to stabilize it. Full service restoration was confirmed by 06:35 US/Pacific as the rollback propagated and servers became healthy.

Google is committed to preventing a repeat of this issue and is taking the following actions:

• Reinforcing rollout processes to include mandatory validation checkpoints.
• Improving alerting systems to monitor critical dependencies more closely.

## Detailed Description of Impact

On Friday, 27 February 2026 between 04:37 and 06:35 US/Pacific, customers accessing Vertex Gemini APIs may have experienced the following:

• Affected Models: All Vertex AI Gemini API models were affected, including gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.5-pro, gemini-3.0-flash-preview, gemini-3.0-pro-preview, gemini-2.0-flash, gemini-2.0-flash-lite.
• Error Experience:
  • PayGo Customers: Experienced primarily 429 Resource Exhausted errors.
  • Provisioned Throughput (PT) Customers: Received 503 Service Unavailable errors.
  • For PT customers, most errors stopped at 06:00. For PayGo customers, most errors stopped at 06:20.
• Geographic Scope: Global endpoint, us-central1, us-east4, and other US regions were impacted.

Preliminary Incident Report

We apologize for the inconvenience this service disruption may have caused. We would like to provide some information about this incident below. Please note, this information is based on our best knowledge at the time of posting and is subject to change as our investigation continues. A final Incident Report with preventative actions will be posted once our investigation is complete. If you have experienced impact outside of what is listed below, please reach out to Google Cloud Support using https://cloud.google.com/support.

Date/Time of the Issue (All time US/Pacific)

Incident Start: 27 February 2026 04:37

Incident End: 27 February 2026 06:35

Duration: 1 hour, 58 minutes

Summary

On Friday, 27 February 2026 at 04:37 US/Pacific, customers using Vertex AI Gemini API models (including Gemini 2.0, 2.5, and 3.0 previews) experienced increased error rates. Impacted services included Google Cloud Support, Agent Assist, the Vertex Gemini API and Dialogflow CX in US regions and the global endpoint for a duration of 1 hour and 58 minutes.

This is not the level of quality and reliability we strive to offer you, and we are taking immediate steps to improve the platform’s performance and availability.

Preliminary Root Cause

This incident was caused by a configuration change to a safety filtering service that supports all Gemini models. This configuration change enabled a code path that interacted poorly with specific requests, leading to service disruption for the safety filtering service. This in turn led to customers seeing overload (429 and 503) errors for their queries.

Google engineers have begun a full root cause analysis and will provide additional information once it is available.

Remediation

Google engineers were alerted to the service disruption via automated alert on Friday, 27 February 2026 04:54 US/Pacific and immediately started an investigation.

Engineers identified the faulty configuration change for a safety filtering service and initiated a rollback to restore the previous stable configuration. Additionally, engineers added more capacity to the service. Full service restoration was confirmed by 06:35 US/Pacific as the rollback propagated and servers became healthy.

Description of Impact

On Friday, 27 February 2026 between 04:37 and 06:35 US/Pacific, customers accessing Vertex Gemini APIs may have experienced the following:

• Affected Models: All Gemini versions were affected, including gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.5-pro, gemini-3.0-flash-preview, gemini-3.0-pro-preview, gemini-2.0-flash, gemini-2.0-flash-lite.
• Error Experience:
  • PayGo Customers: Experienced primarily 429 Resource Exhausted errors.
  • Provisioned Throughput (PT) Customers: Received 503 Service Unavailable errors.
  • For PT customers, most errors stopped at 06:00. For PayGo customers, most errors stopped at 06:20.
• Geographic Scope: Global endpoint, us-central1, us-east4, and other US regions were impacted.

Description
Between Friday, 2026-02-27, 04:36 and 06:45 PST, customers experienced increased error rates when accessing the Vertex Gemini API Global endpoint. The issue impacted API requests to multiple Gemini models.

The incident also caused downstream impact to Dialogflow CX, Agent Assist, Google Cloud Support AI agent, and Customer Experience Agent Studio, which rely on Gemini APIs.

Preliminary analysis indicates the issue was triggered by a recent configuration change. Service was fully restored after the configuration change was rolled back.

We thank you for your patience while we worked on resolving the issue.

Symptom
Customers experienced increased error rates when sending API requests to impacted multiple Gemini models through the global endpoint.