Skip to content

An editorial team and a community building with AI

← All articles

News · Google · 2 min read

Event date: · Published:

Google's Gemini Enterprise Agent Platform gets GA abuse attribution, Agent Gateway preview

Google Cloud's November 11, 2026 release notes bring general availability for per-user abuse attribution in Gemini and Veo, a new CodeMender build, and a limited preview of cross-cloud agent governance via Agent Gateway.

Google Cloud published release notes for its Gemini Enterprise Agent Platform on November 11, 2026, announcing general availability for a request label that attributes policy violations to individual end users, a new CodeMender build, and a cross-cloud agent governance tool in limited preview. The changes are detailed in the Gemini Enterprise Agent Platform release notes.

What changed

  • The safety_identifier request label reaches general availability for Gemini's generateContent and streamGenerateContent calls, as well as Veo's predictLongRunning call.
  • safety_identifier lets multi-tenant apps and API resellers attribute policy violations to individual end users through anonymized IDs, instead of penalizing an entire project for one user's actions.
  • CodeMender reaches version 0.21.0, adding sandboxed build commands, patch-refresh support, read-only host toolchains, isolated caches, and disabled Git hooks for command environment isolation, plus fixes to patch generation failures during sandbox malfunctions.
  • Gemini 4 Carbon is now supported in CodeMender Preview via the --model gemini-4-carbon flag.
  • Agent Gateway, a cross-cloud governance tool, reaches Limited Preview, extending Agent Platform governance (identity, registry, Model Armor policy enforcement, Cloud Logging and Trace) to agents running on non-Google Cloud CNCF-conformant Kubernetes clusters such as EKS, AKS, OpenShift, and bare metal.
  • Agent Gateway is free to use during its preview period.

What it means for AI builders

The two headline changes address practical pain points for teams operating AI agents at scale. Per-user abuse attribution means a single bad actor no longer triggers penalties against an entire project, which matters for multi-tenant applications and API resellers serving many end users through one account. Agent Gateway addresses the opposite problem: applying consistent identity, policy enforcement, and logging to agents no matter which Kubernetes cluster or cloud they run on, rather than limiting governance to Google Cloud infrastructure.

Availability

The safety_identifier label is generally available now for Gemini and Veo calls. CodeMender 0.21.0, including the --model gemini-4-carbon flag, is available in CodeMender Preview. Agent Gateway is in Limited Preview and free to use for the duration of that preview period.


Sources

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.

Google's Gemini Enterprise Agent Platform gets GA abuse attribution, Agent Gateway preview | vibecoding.inc