Securing the AI Enterprise: HiddenLayer Joins the ServiceNow AI Control Tower Alliance

Governance and security belong in the same place

On October 28, ServiceNow named HiddenLayer one of eleven founding members of the Control Tower Alliance, alongside Anthropic, AWS, Cisco, Kong, NVIDIA, Okta, OpenAI, Stripe, and a handful of others. It is a strong group of model providers, hyperscalers, identity, API infrastructure, billing, and security, all building toward the same control plane. HiddenLayer’s piece of that is securing the AI itself. I want to explain why we said yes and what it means for the people running AI programs within large companies.

Here is the problem as I see it from customer conversations. Most enterprises have crossed the line from pilots into production. They have models in cloud accounts nobody remembers provisioning, agents wired into ticketing and CRM systems, and prompts that quietly became business logic. The governance teams have caught up to the inventory problem. What they have not caught up to is the fact that every one of those assets is an attack surface that behaves nothing like the software they already know how to defend.

An inventory does not tell you that a model pulled from a public hub carries a backdoored pickle file. A risk register does not tell you that an agent just accepted a prompt injection buried in a vendor PDF and started leaking records. Governance answers the question of what exists and who owns it. Security answers whether it is going to hurt you. Enterprises need both answers in the same place, and until now they have lived in separate tools run by separate teams.

Why ServiceNow is the right place to put the answers

ServiceNow built AI Control Tower to be the single control plane for enterprise AI: discover it, govern it, secure it, observe it, measure its value. In practice, that is an AI asset inventory covering models, prompts, systems and datasets; a risk and compliance view mapped to authority documents and policies; a security and privacy view that flags dormant and over-privileged agents; and the case workflows that turn a finding into an owner, a deadline and a closed ticket.

The reason this matters is not the dashboard. It is the workflow engine underneath it. ServiceNow is where the enterprise already assigns, escalates, audits, and reports on work. ServiceNow says more than 100 billion workflows run on the platform every year. A security finding that lands in a tool nobody opens is a finding that does not get fixed. A finding that lands in ServiceNow becomes a task with a name on it.

The Control Tower Alliance formalizes that model. Partner solutions go through ServiceNow qualification, certification, and integration testing before customers see them, and select partners build offerings that ServiceNow sells and supports globally. For a customer, that means the integration is tested before they install it, not after. For us, it means we do not have to ask customers to bolt yet another console onto their security stack.

Amit Zavery, ServiceNow’s president and chief product officer, put the premise well in the announcement: “no single solution, no matter how powerful, can see everything.” I agree. Agentic AI touches identity, APIs, models, data and cost all at once, and no one company is going to cover every one of those well. The answer is an ecosystem with a shared control plane, which is exactly what the alliance is: partners build on Control Tower, ServiceNow certifies and integrates the result, and in select cases sells and supports it worldwide.

What HiddenLayer brings that nobody else in the alliance does

Every member of the alliance brings something the others do not, and that is the point. Identity, API gateways, runtime policy, and agent posture are all covered by companies that do them well. What HiddenLayer adds is a focus on the AI itself: the model file on disk, the data it was trained on, the prompts it receives, and the outputs it produces. It is the one layer of the stack that was invented in the last few years, and it is the layer we have spent our whole existence learning to protect.

Our CEO, Chris Sestito, framed it in the announcement as HiddenLayer bringing purpose-built, complete AI security, while ServiceNow brings the enterprise platform and workflow orchestration that turn intelligence into action. That split is the whole thesis. ServiceNow is not trying to build a model scanner or an AI red team, and we are not trying to build a workflow engine. Each side does the thing it is best at, and the customer gets both without stitching them together.

That shows up in four places, and each one maps to a part of Control Tower that today sits empty or partially filled:

• AI Discovery. Control Tower already discovers the AI running on ServiceNow and on the platforms it integrates with, and it is where teams register the AI systems they build. HiddenLayer extends that reach into the cloud accounts, model registries, and ML platforms where models are trained and hosted, and adds the models, datasets, and dependencies that never got registered. Every model we find gets scanned automatically, so the inventory in Control Tower arrives with a security verdict attached rather than a blank field.
• AI Supply Chain Security. Our Model Scanner inspects a model layer by layer for malware, tampering, backdoors, and integrity issues before it reaches production, and produces an AI Bill of Materials and model genealogy so you can prove where a model came from. This is patented technology, and it is the reason a Fortune 100 can pull a model from a public hub without betting the company on it.
• AI Attack Simulation. Automated red teaming that attacks your agents and LLM applications the way an adversary would: jailbreaks, prompt injection, system prompt extraction, policy bypass. Continuously, not once a year before the audit.
• AI Runtime Security. Real-time detection and response on the inputs and outputs of hosted or custom LLMs and agents, including MCP and tool-call traffic. It blocks a malicious prompt before the model sees it and a leaking response before the user does.

Behind all of it is a research team that has spent years finding and disclosing real vulnerabilities in real models and frameworks. That research is what trains our detections. It means our detections are built from attacks we have seen and reproduced ourselves, not from a checklist.

How the integration works

The foundation is the AI Service Graph Connector for HiddenLayer, now available in the ServiceNow Store. This is the first phase of the integration: it imports our AI asset inventory and the security exposures attached to each asset directly into ServiceNow, so customers can start with discovery and supply chain findings right away. As the solution grows, we will bring attack simulation results and runtime detections into the same flow. Here is where the data lands in Control Tower, now and as we expand:

The design principle was simple: HiddenLayer produces the security truth, ServiceNow turns it into work. A model that fails a scan does not get a red badge in our console that a security analyst has to remember to check. It shows up as a risk against a named asset in Control Tower, with an owner and a due date, and it blocks the asset review until someone deals with it.

The same loop runs the other direction. When a new AI system is registered in Control Tower, that is a trigger for HiddenLayer to scan it, red-team it, and put runtime protection in front of it. Governance becomes the front door to security, rather than a parallel process that learns about the agent six months after it went live.

What this means if you run AI at a large company

Three things change for joint customers, and none of them require a new team or a new console.

Your AI inventory is finally complete and finally trustworthy. Control Tower knows about the AI someone registered. HiddenLayer finds the rest in the cloud accounts, notebooks, and registries where it actually runs, and scans it on the way in. The number on the inventory widget becomes the real number.

Security findings become tickets, not slides. A failed model scan, a successful red-team jailbreak, or a blocked prompt injection at runtime lands in the same case workflow your SecOps and GRC teams already use. It gets an owner, an SLA, and an audit trail, and it rolls up into the compliance posture you show the board and the regulator.

You can say yes to agents faster. The reason AI programs stall is not a shortage of use cases. It is that nobody can sign off on risk they cannot see. When every agent in Control Tower carries a HiddenLayer scan verdict, a red-team score, and live runtime protection, the review that used to take a quarter takes a meeting.

The operational payoff is the one I care about most. The security team stops being the department that slows AI down and becomes the department that lets the business turn it on.

Where this goes

What is available today is the foundation, not the finish line. The connector now gets inventory and exposures into Control Tower; each subsequent phase deepens the flows between the two platforms and what Control Tower can do with them. Founding member also means we help set the standards, not just ship to them. Over the next few quarters, we will be working with ServiceNow on shared conventions for agent identity, safety guardrails, and the shape of a security finding, so that what HiddenLayer detects can drive automated containment inside Control Tower, not just a case. We are also extending runtime coverage to agents built on ServiceNow’s own platform and to external agents that integrate with it through Action Fabric.

If you are a ServiceNow customer with AI in production, the connector is now available in the Store. If you are a HiddenLayer customer who has been exporting findings into ServiceNow by hand, that stops now. And if you are somewhere in between, deciding whether your AI program is ready to scale, come talk to us. We will show you what is actually running in your environment, where it stands, and how to get it to a place you can sign off on.

Securing AI has been HiddenLayer’s focus since day one. We are excited to bring it inside the platform the enterprise already runs on.

‍