AI agents are starting to take action in the SOC.
The question is no longer what they can do.
It’s:
Who controls them?
The Missing Layer in AI for the SOC
Most AI innovation in security has focused on capability.
Very little has focused on control.
In a modern SOC, AI agents are not operating in isolation. They are:
- interacting with each other
- executing multi-step workflows
- accessing sensitive systems and data
- operating across teams, environments, and organizations
Without a control layer, this creates real risk:
- Actions taken without boundaries
- Data moving without enforcement
- Workflows executing without oversight
- Limited auditability of what happened and why
This is the gap we set out to solve.
A Milestone for Bricklayer
We’re excited to share that Bricklayer has received a Notice of Allowance for our patent:
“Systems and Methods for Agentic Policy Enforcement”
This represents a foundational step in how AI agents can be safely deployed in enterprise security environments.
At the core of this invention is a simple but powerful idea:
AI agent actions must be evaluated and governed at runtime, before they execute.
This is the difference between automation and controlled autonomy.
What This Enables
Our approach introduces a runtime enforcement system that:
- Intercepts actions taken by AI agents
- Evaluates those actions against policy and context
- Determines whether to allow, modify, or deny execution
- Ensures every decision is consistent, auditable, and enforced
This shifts AI systems from:
- reactive and opaque systems
to:
- controlled, transparent, and enforceable
Why This Matters Now
Security teams are rapidly moving toward AI-assisted and AI-driven operations.
But enterprise adoption requires more than capability.
It requires:
- Governance
- Separation of environments
- Controlled execution of workflows
- Auditability and accountability
Without these, AI introduces risk instead of reducing it.
With the right control layer, it unlocks a new operating model:
AI agents working alongside humans, within clearly defined and enforceable boundaries.
We’ve Seen This Before
What’s happening with AI agents may feel like a new challenge.
In reality, it follows a familiar pattern.
Every major shift toward more distributed, dynamic systems has introduced the same fundamental problem:
How do you control what’s happening across all of those moving parts?
And each time, the industry has responded the same way:
A control layer emerged.
Microservices → Service Mesh
As applications moved from monoliths to microservices, systems became a web of service-to-service interactions.
That created new challenges:
- uncontrolled communication
- inconsistent policy enforcement
- limited visibility
The solution was the introduction of API gateways and service meshes, which:
- intercept requests
- enforce policies
- control how services interact
Cloud → Identity and Access Management
As infrastructure moved to the cloud, the number of users, services, and resources exploded.
It became impossible to manage access manually.
Identity systems emerged to:
- define roles and permissions
- evaluate access requests
- enforce decisions consistently
Kubernetes → Admission Control
In Kubernetes, workloads are constantly being created and modified.
Without controls, anything could be deployed into the environment.
Admission controllers were introduced to:
- intercept requests
- validate or modify them
- allow or deny execution
Networking → Firewalls and Zero Trust
As networks expanded, unrestricted access created significant risk.
The response was:
- firewalls
- segmentation
- zero trust architectures
All designed to ensure that:
every request is evaluated before it is allowed.
The Pattern Is Clear
Across all of these systems, the same pattern emerges:
As systems become more distributed and autonomous, they require a control layer that enforces decisions at runtime.
Bricklayer is building the control plane for AI agents.
This is the shift from automation to controlled autonomy.
AI Agents Are No Different
AI agents introduce the same dynamics:
- multiple independent actors
- continuous interaction
- dynamic decision-making
- access to sensitive systems and data
Without control, this becomes unpredictable.
So the question isn’t:
“What can AI agents do?”
It’s:
“What governs them?”
Identity Is Not Enough
A common assumption is that identity and access management solves this problem.
Identity is critical. It tells you:
- who or what is making a request
- what permissions they have
But identity answers:
Can this happen?
AI systems require something more:
Should this happen, right now, in this context?
That decision depends on:
- prior actions
- workflow state
- data sensitivity
- interactions between agents
Identity is the foundation. But it does not govern behavior.
This Is Just the Beginning
While this Notice of Allowance represents an important milestone, it is only the first step.
Governing AI agents is not a single problem. It spans:
- how agents interact with each other
- how workflows are controlled and adapted
- how environments are segmented and managed
- how policies evolve over time
We are continuing to invest in this space, expanding how multi-agent systems can be governed, controlled, and safely operated at scale.
From Experimentation to Enterprise Systems
We believe the industry is entering a new phase.
The question is no longer:
“What can AI agents do?”
It’s:
“How do we safely operate AI agents at scale?”
The answer isn’t more autonomy.
It’s controlled autonomy.
Building the Foundation for the Future SOC
At Bricklayer, we’re not just building AI agents.
We’re building the system that governs them.
This patent is an important step toward establishing a control layer for AI agents.
Just as service meshes, IAM, and admission controllers became essential in previous system architectures,
AI agents will require this control layer.
The only question is who builds it.
If you’re exploring how to operationalize AI in your SOC with the right level of control and governance, we’d love to connect.
We’ll also be at RSA sharing more about what we’re building.
