Gemini agent platform gains Cisco defenses

AI agents are getting a real enterprise stack now — not just better models, but the boring control layer big companies actually need. That is the story here. Google has rolled out Gemini Enterprise Agent Platform as its main place to build and run agents, and Cisco said on May 11 that AI Defense now integrates with Google’s Agent Development Kit so the same protections can travel from development into production. ### What actually launched? Google’s piece is the platform. Gemini Enterprise Agent Platform is the new umbrella for building, scaling, governing, and optimizing agents on Google Cloud — basically the evolution of Vertex AI into a more explicitly agent-focused system. Cisco’s piece is the security layer. AI Defense now plugs into Google’s ADK, the open-source framework developers use to build tool-using agents, and extends those controls into Google’s managed Agent Runtime. (blogs.cisco.com) ### Why does ADK matter so much? Because ADK is where agent behavior starts. Developers define the model, the tools an agent can call, and the workflow logic there. If security only shows up after deployment, teams end up with the classic handoff problem — one setup in dev, another in production, and a lot of blind spots in between. Cisco is pitching the integration as a way to keep runtime protections consistent across that whole path. (blogs.cisco.com) ### What is Cisco trying to protect against? Mostly the ugly parts of agentic AI. An agent is not just answering questions — it can call tools, hit APIs, touch data, and take actions. That creates a bigger attack surface than a plain chatbot. Cisco has been framing AI Defense around real-time guardrails for agents, plus alignment with security frameworks like OWASP’s LLM top 10, MITRE ATLAS, and NIST’s adversarial ML taxonomy. The basic pitch is simple: if agents are going to act, enterprises need controls that watch the action, not just the prompt. (blogs.cisco.com) ### What is Google trying to become? Google wants one front door for enterprise agents. The new platform bundles model access, agent building, orchestration, DevOps, integration, and security under the Gemini Enterprise Agent Platform name. It also keeps pushing openness — Google’s docs say teams can use ADK or other frameworks, and Gemini models or other models available through Model Garden. So the platform pitch is not “use one model.” It is “run your agent system here.” (blogs.cisco.com) ### Why is this a meaningful pairing? Because it closes a trust gap. Enterprises already know how to pilot an agent. The hard part is letting one loose in a claims workflow, a finance process, or an internal operations system without feeling reckless. Google provides the managed platform. Cisco adds a security story that enterprises already recognize. That combination does not guarantee safe agents, but it does make the buying conversation much easier. That last point is an inference from how both companies are positioning the integration. (cloud.google.com) ### Is this just marketing, though? Partly — of course it is. But it is also where the market is moving. Google announced Gemini Enterprise Agent Platform only about two weeks ago, and Cisco had already been expanding AI Defense on Google Cloud and talking up zero-trust controls for agentic AI. This new ADK integration is the practical connective tissue between those two pushes. (blogs.cisco.com) ### What should enterprises watch next? Watch whether these controls stay framework-specific or become more portable across mixed agent stacks. Also watch whether buyers treat runtime guardrails as optional add-ons or as table stakes. Once agents start using tools and touching production systems, security stops being a feature demo and becomes part of the platform decision. (cloud.google.com) ### Bottom line? The news is not that Google has agents or Cisco has AI security. The news is that the two are stitching those layers together at the exact moment enterprises are trying to move agents from experiments into governed production systems. (blogs.cisco.com)