Cisco AI Defense implementation services, guardrails on every prompt and response.
Cisco AI Defense checks the prompts and responses between your applications and the AI models they call, and blocks what breaks your policy before it reaches the model or the user. Ngenium’s senior engineers design, deploy, tune and validate it in your environment — whether your models are hosted in the cloud or run on your own GPUs — and connect it to your security operations. Services-only: you obtain AI Defense through your Cisco account team or partner; we engineer it.
Everything between a security tool and AI you can trust in production.
AI Defense is a platform, not a switch. Policies have to fit your applications, the gateway has to fit your network and where your models are hosted, and the events have to reach the people who respond. That integration work decides whether it protects on day one.
Know which AI you run, and what it can reach.
We inventory the AI applications, models, data and users in scope — using AI Cloud Visibility where it fits — and review readiness and threats against Cisco’s AI Security and Safety Framework (SAIF), so policy goals are agreed before anything is installed.
Deployed the way your models are hosted.
Cisco’s cloud gateway covers models hosted in the cloud. For models you host yourself, we deploy the hybrid connector on OpenShift or Kubernetes, on your own GPUs or on cloud GPUs in your account, with certificates, DNS and routing handled. Its GPUs are sized in addition to the ones serving your model.
Policies tuned to your applications.
Guardrails against prompt injection, jailbreaks, leakage of personal or sensitive data, unsafe responses, malicious URLs and off-topic use — started in monitor mode, tuned against real traffic, then moved to block with agreed exceptions.
Red-team your models before attackers do.
Algorithmic red teaming of your models and applications, with findings mapped to NIST AI RMF, MITRE ATLAS and the OWASP Top 10 for LLMs. The report supports your security reviews and gives your team evidence to use toward ISO/IEC 42001 or ISO/IEC 27001 — evidence, not certification.
Agent tool traffic and unsanctioned AI apps.
Visibility and policy for agent and tool traffic (MCP) and, combined with Cisco Secure Access, discovery of the third-party AI apps your employees use, with the unsanctioned ones restricted.
Events your security team can act on.
Allowed and blocked traffic logged with the rule behind each decision, wired into your SOC tooling, including Splunk, with runbooks and knowledge transfer — or a hand-off to Port53 for 24x7 managed security operations.
Running it on Cisco AI PODs? See our AI PODs deployment services →
Assess, design, deploy, operate: one accountable team.
From a first conversation to a protected application, the same senior engineers carry the work. After go-live, AI Defense can move into managed security operations through our strategic partnership with Port53.
30 days to prove it on one of your applications.
Not ready for a full rollout? Ngenium runs fixed-scope 30-day evaluations. The 30-Day Proof of Concept answers “does it work in our environment?” with one application protected end to end and the event log as evidence. The 30-Day Proof of Value adds success criteria agreed at kickoff, one AI Validation run, a scorecard and an executive readout with a recommended path to production. Both are arranged through your Cisco account team.
AI Defense inside a full stack, working within days.
At a major U.S. healthcare provider, a multi-vendor Secure AI Factory deployment had stalled for months. Ngenium engineers delivered the full stack — UCS with NVIDIA GPUs, OpenShift, Isovalent, AI Defense and Splunk — working and observable within days.
Why teams bring in Ngenium.
- Senior engineers only — the engineer who designs your AI Defense deployment is the one who installs and tunes it.
- Services-only — we never resell hardware or software and carry no product margin; you obtain AI Defense through your Cisco account team or partner, and we engineer it.
- Cisco depth, multi-vendor fluency — AI Defense with Secure Access, Splunk, Isovalent, OpenShift and NVIDIA, on Cisco AI PODs or your own infrastructure.
- Regional reach — senior engineers across North America, Europe and Latin America.
About Cisco AI Defense implementation.
What is Cisco AI Defense?
Cisco describes AI Defense as security for AI applications: it discovers AI workloads and applications, validates models with algorithmic red teaming, protects applications at runtime with guardrails against threats such as prompt injection and data leakage, and manages access to third-party AI apps. Cisco references NIST, MITRE ATLAS and the OWASP Top 10 for LLMs.
Does Ngenium sell AI Defense?
No. Ngenium is services-only. You obtain AI Defense through your Cisco account team or a Cisco partner; Ngenium designs, deploys, tunes and validates it. Cisco channel partners also bring Ngenium in to deliver the engineering for their customers, including white-label engagements under NDA.
Where does it run: Cisco’s cloud or our own GPUs?
It depends on where your models are hosted. Cisco’s cloud gateway covers models hosted in the cloud, but it cannot protect models you host yourself. For those, Cisco documents a hybrid deployment: a connector that runs Cisco’s guardrail models on GPUs in your OpenShift or Kubernetes cluster, on-premises or in your own cloud account, with policies written and events read in Cisco’s cloud console. Those GPUs are in addition to the ones serving your model. We review data handling with your security and privacy teams as part of the design.
How long does it take?
Our fixed-scope 30-day evaluation takes thirty days from readiness call to readout, for one application. A production rollout depends on how many applications and models are in scope, where they are hosted and how much policy tuning your traffic needs. A 30-minute scoping call gives you an honest estimate.
Does it work with Cisco AI PODs, Secure Access and Splunk?
Yes. Cisco publishes a validated design for AI Defense on Cisco AI PODs, and Ngenium deploys both. Combined with Cisco Secure Access, AI Defense also helps with shadow AI by finding the third-party AI apps your employees use and restricting the unsanctioned ones. Splunk customers can use their Splunk data for additional insight, and events can feed your SOC.
Planning to protect an AI application?
Talk to the senior engineers who would deploy it — not a salesperson.