ModelOp adds operating system level policy enforcement for AI agents
A new integration with NVIDIA OpenShell allows companies to control exactly what AI agents can read, write, or run at the system level.
The short version
- ModelOp now enforces AI policies at four levels: process, execution, operating system (OS), and network.
- The new integration uses NVIDIA OpenShell, an open-source secure runtime for AI agents.
- The system uses deny-by-default, kernel-level controls that run outside of the agent's own process.
- The platform connects with existing tools like Microsoft Agent 365, Amazon Bedrock AgentCore, and Kong gateways.
Why it matters
For companies deploying autonomous AI, this moves governance from 'paper compliance' to actual technical restrictions. It ensures that an agent cannot bypass security measures by finding unintended ways to access files or move data once it is live.
Closing the governance gap
As companies rush to deploy autonomous AI agents, a significant security gap has emerged between approval and production. An AI agent might be fully approved and compliant on paper, yet it can still take actions that violate corporate policies once it begins operating. ModelOp is addressing this by launching Full-Stack AI Policy Enforcement to manage these risks continuously.
The company argues that traditional governance often ends at the deployment phase, which only documents what a company hopes an agent will do. Because agents can execute code, access sensitive files, and communicate with external systems on their own, they require active restrictions. ModelOp aims to bridge the gap between high-level risk assessments and the actual behavior of the software in a live environment.
By centralizing these rules in a single system of record, the platform ensures that policies are not just static documents. Instead, they become active constraints that govern how an agent behaves at every stage of its lifecycle. This approach is designed for large enterprises, including Fortune 500 companies, that need to industrialize AI while maintaining strict oversight.
Four levels of enforcement
ModelOp identifies four distinct levels where AI policy must be enforced to be effective. These include the process level, which covers the end-to-end lifecycle of an agent, and the execution platform where the agent actually runs. It also includes the network level, which governs how the agent connects to external systems and internal APIs.
The newest addition to this framework is the operating system (OS) level. This is where the platform now enforces what an agent can read, write, run, and send. Without OS-level controls, an agent might find a way to bypass an AI gateway by running its own code to reach a restricted destination. ModelOp uses kernel-level controls to stop these actions from the outside.
The platform integrates with several existing enterprise tools to provide this multi-layered coverage. It works alongside Microsoft Agent 365 and Amazon Bedrock AgentCore for execution, as well as Kong for network gateways. By connecting these disparate layers, ModelOp ensures that there are no gaps where an agent can slip through unmanaged permissions.
How the NVIDIA integration works
The new integration specifically utilizes NVIDIA OpenShell, which is an open-source secure runtime designed for AI agents. ModelOp turns the sandbox permissions of OpenShell into governed policies tied to specific business purposes. This means permissions are no longer isolated technical settings but are linked to an agent's owner, risk tier, and approved version.
The system operates through a three-step process involving management, enforcement, and observation. First, security and IT teams set global policies in ModelOp, while solution owners add specific requirements for their use cases. ModelOp then records these policies for every version of an agent, ensuring that permissions carry over correctly during updates.
During the enforcement phase, ModelOp registers OpenShell gateways and tracks which agents are running in which sandboxes. If OpenShell blocks an action, the system routes the exception through a review workflow. This allows the right reviewers to see the agent's owner and risk tier before they decide whether to approve a policy change. As Jim Olsen, CTO of ModelOp, noted, "Each enforcement point sees one slice of what an agent does."
Impact on enterprise security
For Chief Information Officers and Chief Information Security Officers, this tool provides a way to scale AI agents more safely. It moves the responsibility of security from individual teams to a centralized command center. This allows for consistent enforcement across different departments that might otherwise use different tools or have no shared record of policies.
The platform is built to work with existing investments in governance, risk, and compliance (GRC) and IT service management (ITSM). Rather than replacing these systems, ModelOp connects to them to provide a unified operating layer. This helps regulated institutions, such as banks and healthcare organizations, accelerate their AI adoption without sacrificing security.
By providing a system of record for every version of an agent, the platform also simplifies auditing. Every decision made by the system—including blocked actions and approved policy changes—is recorded for review. This creates a transparent trail of how AI agents are behaving and how their permissions are being managed over time.
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
ModelOp's move into OS-level enforcement signals a shift toward more rigorous technical standards for autonomous systems. By using deny-by-default controls, the platform ensures that agents can only perform the specific tasks they were designed for. This reduces the risk of
The path toward industrial AI
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by The National Law Review · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.