Google Cloud revealed various AI-driven security features for its platform, with AI agents and identity key themes among the announcements.

On the identity front, the forthcoming Agentic IAM service is tailored around AI agents, AI models that perform tasks autonomously for enterprises, processing data and continuously adapting based on those results.

Google Cloud's IAM service is designed to automatically provision agentic identities across all agent development runtimes, while supporting a range of credential types, authorization policies, and full end-to-end observability.

The updates come as agentic AI increases its foothold in the networking space, with Cisco recently suggesting to SDxCentral that the impact of numerous AI agent deployments could strain network bandwidths equivalent to “80 billion” users.

With the Agentic IAM tool, Google Cloud Platform (GCP) users will be better able to authenticate and authorize AI agents amid the potential traffic onslaught, a feature that echoes recent software updates by security firm CrowdStrike.

Following Oracle with MCP

Elsewhere in GCP, the Security Command Center (SCC) will see new capabilities for model context protocol (MCP).

This open standard, which was introduced by AI firm Anthropic, aims to standardize the way AI systems – in particular large language models (LLMs) – integrate and communicate with external data sources, tools, and systems. Such integration can help developers query data, generate reports, run vector queries, and perform data handling operations.

Details on the move were not detailed by Google, but hyperscale rival Oracle recently integrated MCP into its Database platform. This saw MCP tools provided to the AI assistant and its LLMs, allowing connection to an Oracle Database and executing SQL queries and scripts.

The networking side could potentially see Oracle Cloud Infrastructure (OCI) or GCP administrators filtering and monitoring MCP activity, keeping sessions separated using network segmentation, and sending logs from the MCP server to central security platforms for real-time tracking.

What’s new in SCC

Google Cloud also previewed updated services for its SCC, such as the Compliance Manager tool, which now includes AI controls to automate AI workload compliance through built-in baselines, AI-specific controls, reporting, and continuous monitoring.

When these are enabled, the platform can ensure that model endpoints sit behind private connectivity mechanisms to prevent deployments onto open public IPs. This allows for network boundaries without the creation of bespoke rule sets, and any attempt to break those boundaries can be automatically logged as a compliance breach.

In addition, the Data Security Posture Management (DSPM) governance tool integrates natively with BigQuery Security Center, enabling users to monitor data security and compliance directly within the BigQuery console, eliminating the need to switch between platforms.

Where earlier releases kept DSPM findings in a separate dashboard, the new version embeds them natively into the BigQuery console, much like Microsoft Azure's Synapse Analytics, which is embedded within Microsoft Defender for Cloud.

For GCP users, the BigQuery integration means a dataset that slips into public status is flagged to the user, with information on who can reach the data, much like when a public port is flagged in a firewall setting.

Patching the exposure inside BigQuery potentially allows the same engineers who manage firewalls find the problem at source, rather than tracing it through multiple tools, in an example of the increasingly popular shift-left security trend, which sees network-security checks occur earlier in workflows.