Google Cloud expands security controls for AI agents

Putting agents into production means managing their credentials and boundaries as well as their prompts.

What happened

Google Cloud announced expanded agent-security capabilities on 4 November. Auth Manager is generally available for centralised credential management. Agent Identity JWT support and custom topic controls in Model Armor are in preview.

CodeMender, also in preview, gained isolation and scanning options. The announcement contains a mix of generally available capabilities and previews.

Why it matters

An agent may need access to several systems to complete a task. Credential handling, limits on that access and controls around code execution become part of the deployment, alongside the model's ability to produce a useful answer.

The bigger picture

Availability labels matter for teams deciding what they can use today. The announcement should not be flattened into a claim that every feature is ready for general deployment. The practical value will depend on how the controls fit the environment and whether teams configure them effectively.