WESTGATE DATA SCIENCE / RUNTIME CLEARANCE

Churchill's Protocol.What runs is what you approved.

Churchill verifies what is actually running against the version your governance board approved. Every time it runs, continuously while it runs. Anything else is blocked at the kernel boundary before it executes.

Verified continuously against the approved package.

Even with root, nothing executes that the Change Advisory Board did not approve.

THE LANDSCAPE

Where the jewels live. How threats reach them.

The protected applications that hold the data.

Where the crown jewels live

The systems that hold the data and run the regulated business.

  • Banking cores
  • Settlement engines
  • Payment rails
  • Electronic health records
  • Pharmacy and lab systems
  • AI agent runtimes
  • Industrial control systems
  • ERP and supply chain backbones
  • Mainframe partitions running the regulated business
Once they're in, anything can happen

Upstream verification has a known limit.

Build pipelines sign whatever the build infrastructure produces. Identity layers verify whoever has the credential. Both operate upstream, before code runs on the system that holds the data.

Once an attacker is past the upstream layer (through a compromised build, a stolen credential, or an autonomous AI agent with delegated permissions) anything can happen on the system that holds the value.

Churchill verifies what is actually running on the system that holds the data, at the moment it runs. After upstream verification. After the attacker is already inside. Even with root.

HOW IT WORKS

Setup follows your governance. Then Churchill enforces continuously.

After Churchill is downloaded, a guided UI walks your team through two governance steps. Once complete, the system runs autonomously until the next release or patch.

01
Install

Controlled deployment

Churchill installs on your protected systems through a controlled download. No pipeline changes. No modifications to your build infrastructure.

02
Quorum

Establish governance

The Churchill UI prompts you to designate a minimum of two quorum members from your Change Advisory Board. We recommend pairing your release decision-maker (Release Manager or VP Engineering) with your executive sponsor (CISO or CTO).

03
Lock

Quorum seals the application

Your quorum selects the application to protect and clicks enter. The approved version is captured as a cryptographically signed snapshot: executables, configurations, libraries, scripts. That snapshot now defines what is allowed to run, verified in real time at full speed. Anything outside it is diverted to the kernel-level mirror envelope before it executes and preserved as forensic evidence.

Platform Requirements Linux kernel 5.7 and higher. Validated on IBM LinuxONE. Compatible with Linux on x86 and Power, containers, virtual machines, and bare metal.
ONGOING OPERATION

Two signatures per release. Seconds each.

You already approve every deployment, patch, and rollback, usually through emails and loose decisions. Churchill replaces that with one electronic signature: sign once to take the application down, once to bring it back. Seconds each, 2 a.m. or 5 p.m., on your schedule and no one else's. Nothing else about maintenance, patches, or releases changes.

During releases and patches

Two signatures. Seconds each.

Nothing about your release process changes. CI/CD builds, governance approves. At the moment of approval, Churchill prompts the quorum you already designated to sign, in seconds, at 2 a.m. or 5 p.m., on your schedule.

  • Sign to take the application down.
  • Churchill re-seals the new approved version.
  • Sign to bring it back. Enforcement resumes against the new state.
~2 seconds per release cycle
Between releases

Nothing to manage.

Churchill runs autonomously against the sealed application. The system is protected continuously. The CISO and CAB receive intelligence on attempts, on their own review schedule.

  • No ticket queue to triage.
  • No daily review burden.
  • No additional FTEs required.
  • No alert-fatigue cycle.
Protection + intelligence. No damage.

The security itself does not create the next operational problem. We thought this through.

THREAT INTELLIGENCE CAPTURED BY DESIGN

The byproduct is threat intelligence.

When Churchill blocks an unapproved change, the action does not just stop. It is diverted into a kernel-level mirror envelope. The attacker believes they succeeded. They keep working. They reveal their entire playbook, in real time, against a system that does not exist.

What happens, step by step
01 The Insider

Attempts an unapproved change

Could be a privileged employee. Could be a credential-theft attacker. Could be an autonomous AI agent. Even with root.

02 Churchill

Blocks pre-execution at the kernel

The action is verified against the quorum-approved application. It does not match. It does not execute on production.

BLOCKED · >2ms
03 Mirror Envelope

Captures the full playbook

The action is diverted into a kernel-level mirror. The insider believes it succeeded. They keep working. They reveal everything.

RECORDING
Production: untouched.

The protected application keeps running. No outage. No remediation queue. No incident response window. No ticket queue.

The CISO: gets the evidence first.

Pre-execution blocked transaction log. Recording of the methodology. Behavioral telemetry of the mirror execution. Before the insider knows they were identified.

Churchill holds while you assess the insider, before they suspect you're onto them. Even with root, the insider cannot unilaterally change the system.

FULL COMPARISON

How Churchill compares to the rest of the stack.

Where Churchill operates relative to every other layer of the security stack.

Comparison of security stack layers by tooling, mechanism, and timing and scope, showing where Churchill's runtime clearance fits.
Layer Tooling Mechanism Timing & Scope
Application EDR, SIEM, RASP, WAF Observes application activity and network traffic. Recognizes known attack patterns from what it sees. After execution. Generates alerts for review.
Identity IAM, sudo, RBAC Verifies who is requesting access by checking credentials. Before access. Identity-based decisions.
File integrity FIM Detects when monitored files have been modified on disk. After modification. Alert-based.
Binary execution Whitelisting Checks each executable's digital fingerprint against an approved list before it runs. Pre-execution. Each executable checked individually.
OS hardening Linux Security Modules Enforces predefined rules about what processes are allowed to do on the operating system. Rule-based. At specific operating system decision points.
Runtime clearance Churchill Verifies the protected application against the version your Change Advisory Board approved (recommended members: CTO, CISO, Release Manager). Continuously, while it runs. Pre-execution. Full application. Every action. Requires governance, by design.
Firmware Secure Boot, TPM Verifies the system's boot sequence has not been modified, using hardware-based trust. At system startup only.
THE ARCHITECTURE

Three layers. One gate.

The Watcher confirms Churchill is operating. Churchill verifies what executes. The Protector anchors Churchill in the kernel, below the reach of user-space or root.

See the full architecture
NEXT FROM WGDS

One mission. Two surfaces.

Churchill applies runtime clearance to the execution surface. Digital Detectives applies policy as code to the data surface. Built on the same mission. Launching next.

Preview Detectives
How threats reach the jewels

The endpoint is the entry. The server is the target.

The Windows endpoint is the entry. The Linux server is the target.
A common CISO question

"Aren't most attacks coming through Windows endpoints, not Linux servers?"

We agree. The endpoint is the landing site. The Linux server is the damage site.

The endpoint is where phishing lands. The file server, the database, the application server, and the AI runtime are where encryption, lockup, exfiltration, and pivot actually happen.

EDR does the real upstream work, and Churchill doesn't displace it. Churchill protects the critical Linux systems downstream where the actual loss occurs, plus Windows CE for embedded and critical control environments.

Endpoint defense and identity tools belong on the Windows endpoint. Churchill belongs on the system the attacker is trying to reach.

Built and validated on IBM LinuxONE.

Linux kernel 5.7 and higher. Compatible with Linux on x86 and Power, containers, virtual machines, and bare metal. WestGate Data Science is an IBM Technology Partner.

ANTHROPIC Cyber Verification Program

Adversarially tested through Anthropic's Cyber Verification Program.

Mythos engagement: 29 documented sessions, 11.3 cumulative hours, 406,433 enforcement decisions, multiple sessions starting with full root. 100% containment. Zero data exfiltrated.