As organisations accelerate AI adoption, security leaders are being forced to confront a new reality. AI workloads introduce fundamentally different risks to traditional applications, from model manipulation and prompt injection to data leakage across complex agentic workflows.
This week at Google Cloud Next 2026, Cisco announced a major expansion of Cisco AI Defence, extending native support to Google Cloud alongside existing coverage for AWS and Microsoft Azure. The move positions Cisco AI Defence as a truly multi cloud AI security platform, designed to protect modern AI applications wherever they run.
This announcement reflects a growing consensus across the industry. As AI systems become more autonomous and distributed, security can no longer be bolted on after deployment.
What is Cisco AI Defence?
Cisco AI Defence is an enterprise security platform built specifically to protect AI models, applications and agentic workflows across their full lifecycle. Cisco originally unveiled AI Defence in January 2025 as a response to the growing gap between AI innovation and security readiness.
Unlike traditional security tools that focus on infrastructure or endpoints, AI Defence is designed to address the unique behaviours and risks introduced by AI systems.
At its core, Cisco AI Defence provides:
Across more than 200 AI security and safety subcategories to proactively identify weaknesses in models and applications
That applies bi‑directional guardrails inline to AI interactions, including prompts, responses and tool usage, without requiring code changes
Including Retrieval Augmented Generation pipelines, tool‑calling agents and model orchestration frameworks
Across AI assets such as models, agents, data sources and tooling to support governance and compliance
What Has Changed: Google Cloud Support Announced at Cloud Next
The latest announcement extends Cisco AI Defence to Google Cloud, making it the third major hyperscaler supported alongside AWS and Azure.
For organisations building on Google Cloud services such as Gemini Enterprise Agent Platform, Vertex AI or Kubernetes‑based AI workloads, this means they can now apply consistent AI security controls across all three major public cloud environments.
Key elements of the Google Cloud integration include:
- Inline enforcement through Google Kubernetes Engine Service Extensions, enabling runtime protection without modifying agent or model code.
- Support for securing agentic AI workflows via integration with Google Cloud Agent Gateway
- Optional VPC‑based deployments that ensure prompts, responses and model interactions remain entirely within the customer’s Google Cloud environment
This expansion aligns closely with Google Cloud’s own focus on agentic AI security, which was a central theme at Google Cloud Next 2026.
Why Multi Cloud AI Security Matters Now
Most enterprise AI strategies are already multi cloud, whether by design or necessity. Models, data sources and agents are often distributed across platforms, creating inconsistent controls and visibility gaps.
The security implications are significant. According to the 2025 Cisco Cybersecurity Readiness Index, 86% of organisations experienced an AI‑related security incident in the past 12 months, yet fewer than half believe they have the internal expertise required to secure AI effectively.
Additional findings from the same research highlight the scale of the problem:
- Only 4% of organisations globally have reached a mature level of cybersecurity readiness in the AI era
- 49% of leaders say employees do not fully understand AI‑related cybersecurity threats
As AI agents gain greater autonomy and access to enterprise systems, the blast radius of a single misconfiguration or compromised model increases dramatically.
Securing Agentic AI and RAG Pipelines
One of the most significant challenges addressed by Cisco AI Defence is the rise of agentic AI. These systems do not simply respond to prompts, but take actions, call tools and interact with multiple data sources.
Cisco AI Defence is built to secure this new class of workload by:
- Detecting and blocking prompt injection and model manipulation attempts at runtime
- Preventing unauthorised tool usage and data exfiltration in agent workflows
- Applying consistent policy controls across Retrieval Augmented Generation pipelines, reducing the risk of sensitive data leakage
This capability is particularly relevant as more organisations move AI applications from experimentation into production environments.
A Platform Built for Enterprise Reality
What differentiates Cisco AI Defence is its emphasis on enterprise‑grade deployment, governance and scale. Rather than acting as a point stand‑alone tool, AI Defence is designed to integrate with existing security, networking and cloud operating models.
By supporting AWS, Azure and now Google Cloud, Cisco enables organisations to apply a single AI security framework across heterogeneous environments, reducing complexity while improving control.
As Cisco notes, AI adoption cannot be sustainable without security being embedded into the foundation of AI operations
The expansion of Cisco AI Defence to Google Cloud is a clear signal that AI security is becoming a mainstream enterprise priority. With AI‑related incidents already impacting the majority of organisations, the question is no longer whether AI needs to be secured, but how quickly and consistently this can be done across multi cloud environments.
For organisations investing in agentic AI, RAG pipelines and large‑scale AI workloads, this announcement removes a key barrier to adopting a unified security approach.
Ready to Secure AI Across Your Cloud Environments?
AI innovation is moving fast. Security needs to move faster.
If you are exploring AI workloads across AWS, Azure or Google Cloud, now is the time to assess how prepared your organisation really is. Speak to us about designing an AI security architecture that protects models, data and agentic workflows without slowing innovation.
Start a conversation with our security specialists to understand what enterprise‑grade AI defence looks like in practice.