The Risk Is Already Right here
In February 2026, safety researchers found that the SmartLoader malware operation had cloned a reputable Mannequin Context Protocol (MCP) server, one which connects synthetic intelligence (AI) assistants to Oura Ring well being information and submitted a trojanized model to reputable MCP registries. The attackers spent months creating pretend GitHub accounts, manufacturing contributor credibility, and constructing a community of fraudulent forks. Builders trying to find a well being information integration would have put in credential-stealing malware with no apparent warning indicators.
This was not proof of idea. It was a stay assault that efficiently poisoned registries utilized by actual builders. And it represents a category of provide chain danger that grows with each new MCP server, device integration, and AI agent that enterprises deploy.
Most enterprise AI agent platforms depart it to particular person builders to vet each integration manually. That strategy doesn’t scale when the assault floor is increasing sooner than any group can assessment. At present, we’re saying that Cisco AI Protection is constructed straight into Cisco Agent Builder, making it the primary enterprise AI agent platform with native safety at each lifecycle stage. From scanning third-party integrations earlier than they attain builders, to inspecting each agent execution in actual time, safety is now a part of the platform itself.
What Is Cisco Agent Builder?
Agent Builder is one of many capabilities being introduced inside Cisco Cloud Management Studio, the brand new a part of Cisco Cloud Management the place enterprises flip their third-party instruments, operational data, and workflows into AI brokers that may act throughout their complete atmosphere. It’s the place three forms of work occur:
Connecting third-party instruments. Enterprise ITSM, monitoring, DNS, identification, and alerting instruments join by way of native integrations. As soon as linked, a device’s information and actions develop into accessible to AI brokers working throughout the platform, together with inside Cisco AI Canvas in Cloud Management.
Constructing customized AI brokers. Brokers are in-built a guided interface that doesn’t require deep coding abilities. Builders describe what the agent ought to do, connect the instruments and data it wants, check it, and publish it by way of a versioned lifecycle.
Encoding operational data as abilities. Runbooks, commonplace working procedures, compliance requirements, and remediation procedures are uploaded and transformed into reusable abilities that brokers can name when the work calls for it.
What’s added in Agent Builder doesn’t keep there. Related instruments, customized brokers, and operational abilities floor throughout Cisco Cloud Management and its workspace, AI Canvas, the place human operators and AI brokers examine and resolve points collectively.
The platform makes it easy to construct and deploy AI brokers. The tougher query is: how do you safe them?
Constructed In, Not Bolted On
Agent Builder solutions that query by embedding safety at each stage of the agent lifecycle by way of Cisco AI Protection working collectively as one built-in safety layer.
Earlier than an integration is on the market: Cisco AI Protection scans each third-party MCP server’s code, it’s configuration like device definitions, and information flows for vulnerabilities, malicious habits, and provide chain dangers. Integrations that fail scanning are by no means proven to builders. A trojanized MCP server just like the SmartLoader Oura clone can be blocked earlier than any builder ever encounters it.
Earlier than an agent is totally constructed: AI Protection mechanically scans agent configurations for immediate injection patterns, information leakage dangers, and coverage violations each time a builder saves a draft.
Earlier than a ability reaches manufacturing: AI Protection Ability Scanner validates ability directions and uploaded ability markdown for embedded adversarial content material and delicate information publicity.
Throughout each execution: AI Protection inspects each Massive Language Mannequin (LLM) name and each device invocation in actual time. Consumer inputs are checked for threats like immediate injection and jailbreak makes an attempt earlier than they attain the mannequin. Agent responses are checked for information leakage (personally identifiable data, credentials, inner community addresses) earlier than they attain the person. Coverage Studio is used to mechanically create customized guardrails that block actions after which log the occasions within the execution hint for full auditability.
Builders don’t configure any of this. It occurs mechanically for them in Cloud Management. They construct brokers, get a inexperienced checkmark, and deploy. Safety runs invisibly at each gate.
Picture: An agent in Cisco Agent Builder secured by AI Protection.
One Firm. One Platform. One Safety Posture.
This isn’t a market integration or a partnership announcement. That is Cisco’s personal AI safety capabilities, defending Cisco’s personal AI agentic platform. AI Protection is constructed by Cisco AI and embedded straight into Agent Builder.
The sign that sends is simple: Cisco trusts these safety merchandise sufficient to place them contained in the platform it sells to enterprises. If they’re trusted by Cisco, they’re able to ship that very same AI safety excellence on your group.
Examine this to the choice. Most AI agent builders require enterprises to bolt on third-party scanning instruments, configure runtime safety individually, and handle safety insurance policies throughout disconnected techniques. When one thing goes fallacious, there are integration gaps and finger-pointing. When the platform and the safety come from the identical firm, there may be one group to name and no seams to use.
The Agent Period Calls for Native Safety
The MCP ecosystem is rising quickly, and risk actors have observed. Conventional belief indicators like GitHub stars, fork counts, and contributor lists can now be systematically fabricated. Guide assessment doesn’t preserve tempo.
Cisco Agent Builder, secured by Cisco AI Protection, provides enterprises a platform the place each integration is vetted, each configuration is checked, and each execution is inspected, mechanically, earlier than something reaches manufacturing.
Be taught extra about Cisco AI Protection and Agent Builder in Cisco Cloud Management. Should you’re becoming a member of us at Cisco Stay Las Vegas this week, you may go to our respective cubicles to see these firsthand.
Some merchandise or options described could also be in varied levels of growth and provided on a when-and-if accessible foundation.
