Google Cloud launches new security tools for AI agents

Google Cloud has introduced new security products and updates for organisations using autonomous AI agents. The changes include updates to CodeMender, new identity controls for agents and closer integration with Wiz.

The rollout focuses on three areas: securing software development, tightening controls around production workloads, and linking threat detection and response more closely across Google's security products and Wiz's platform.

One of the main additions is CodeMender, Google Cloud's code security agent, which remains in preview. It is designed to help organisations find and fix software vulnerabilities more quickly by validating flaws and generating tested patches before attackers can exploit them.

CodeMender now supports isolated verification and exploit simulation in either an operating system sandbox or a virtual machine, alongside human checks for local write and command execution. Google has also added more granular scanning options, allowing customers to scan a folder, a full repository or only files changed since the last scan.

Google said hundreds of customers have used CodeMender since the preview opened in July. It also highlighted recent work by Wiz using Gemini 4 Argon with other models to identify and help remediate core PHP vulnerabilities affecting 70% of websites.

In that case, Google said Gemini 4 Argon helped find the vulnerability, develop a likely exploit path, detect attacks and validate the patch.

Industry reaction from early users focused on speed and oversight. "As adversaries attack at machine speed, Google's Fairwind Program helps us keep pace with our application security. Our early evaluation of CodeMender shows the potential to reduce vulnerability discovery time by over 95%, while keeping expert validation and remediation under human oversight," said Ivan Fontarensky, CTO of Cybersecurity Services at Thales Group.

IDC also linked the launch to mounting remediation pressure on security teams. "IDC research finds that nearly 40% of enterprises add vulnerabilities faster than they can remediate them. Detection without remediation only adds to developer workloads," said Katie Norton, Research Director of Cloud Security at IDC. "CodeMender solves this by pairing exploitability validation with deployable code fixes, driving the continuous, verified patching required as exploitation timelines compress."

Identity controls

Google Cloud also introduced controls aimed at the infrastructure autonomous agents use once they move into production environments. It said attackers are increasingly targeting credentials, data boundaries and integrations that connect agents to enterprise systems.

Auth Manager is now generally available and is designed to centralise credential management in a Google-managed vault governed by identity and access management policies. Google said the service automates OAuth 2.0 and PKCE flows and refreshes user-delegated tokens, reducing reliance on repeated user consent and lowering the risk of credential theft.

Agent Identity JWT support has also entered preview. Google said it issues standard OIDC tokens so agents can authenticate across Amazon Web Services, Microsoft Azure, on-premises back ends and MCP servers without relying on static, long-lived API keys.

The company is also extending identity-based controls to network boundaries. VPC Service Controls ingress and egress rules now support IAM roles, which are generally available, while Workload Identity Federation for Google Kubernetes Engine is in preview.

Google said these changes are intended to help security teams apply more granular perimeter controls across enterprise GKE clusters without having to manage large numbers of manual exceptions.

Workload security

Further product changes cover content and network inspection. The Model Armour custom topic detector, in preview, allows organisations to define prompt-based rules to identify whether agent interactions stay within approved topics. Google said the feature is aimed at sectors including healthcare, insurance and retail.

On the network side, Cloud NGFW advanced malware sandboxing is now generally available and uses Palo Alto Networks WildFire. Google has also made Cloud NGFW project-level resources generally available and added Cloud Armour managed rules in preview, supported by threat intelligence from Thales Imperva.

Wiz links

Google is also expanding its relationship with Wiz as customers try to manage security across cloud environments and development pipelines. Wiz Defend and Google Security Operations now share a common data model and support bi-directional synchronisation.

The integration also includes a dedicated Wiz Content Pack in Google Security Operations, Wiz Blue Agent investigation analysis in Google Security Operations, and an integration between Wiz Defend and Google Threat Intelligence.

Wiz AI Static Application Security Testing, or Wiz AI SAST, has also entered preview. Google said the tool uses agentic reasoning to analyse application behaviour and identify business logic flaws, with findings correlated against infrastructure context in the Wiz Security Graph.

A joint project between Google DeepMind and Wiz, called Scan for Good, has also started work with public service, critical infrastructure and non-profit organisations. Google said the initiative used Argon in its first month to help more than 50 public service organisations and critical infrastructure providers find and remediate critical exposures.