Score your stack.

Select the statement that best describes your current infrastructure for each of the four CRC pillars. Your Minimum Surface Score is a structural posture screen for the AI-era threat model.

The MSS score is not a graph-risk score. CRC reachability and federation constraints are deliberately not collapsed into the 0–16 number. Systems with the same structural score can have very different traversal depth, edge authority, cross-surface data movement, and blast-radius characteristics. Measure the execution envelope separately →

Governed Server Boundary prerequisite: This assessment applies only to governed server deployments under formal organizational control. If AI agents with computer-use capability can reach your governed boundary from unmanaged devices, this assessment is not applicable.

MSS Score

-- / 16

Select all three pillars

01 Consolidate

How many integration boundaries exist between your regulated workflow applications?

Uninspected

Multiple enterprise applications, integration boundaries unowned and uninventoried. Nobody is formally responsible for what happens at the API layer between systems.

Inventoried

Integration boundaries are documented and ownership is assigned. Data flows are mapped. The gaps are known even if they are not yet defended.

Controlled

Boundaries are owned, monitored, and traffic is logged. All data flows produce observable records. API contracts are enforced and audited.

Consolidated

The majority of regulated workflow runs on a single governed platform. Integration boundaries between external systems are minimized and each is a conscious architectural decision.

Minimal

Single governed platform. Zero unowned boundaries. Every data flow produces an immutable, attributed audit record. There are no integration gaps to chain across.

02 Reduce

What is present in your production execution environment beyond the application itself?

Uninspected

General-purpose OS with full package universe. Shell present and accessible in production. No inventory of running services. Default kernel with all subsystems active.

Inventoried

Running services are documented. Unnecessary services identified and disabled. OS hardening applied per a recognized benchmark (CIS, STIG). Shell access is restricted and logged.

Controlled

Containerized isolation. No package manager in production. Minimal base image. Shell access requires explicit break-glass procedure with full audit trail.

Reduced

Purpose-built OS image. No shell binary present in production. Read-only root filesystem. Signed boot chain. Only the components required to run the application are included.

Minimal

TPM-attested measured boot. Immutable filesystem. Application is the OS -- the execution environment contains exactly the application and nothing else. Kernel surface minimized to required syscalls only.

03 Close

What can reach your production systems from outside the governed boundary?

Uninspected

Public-facing services, unrestricted egress, ports open by default. No formal network inventory. External parties can reach production systems without explicit authorization.

Inventoried

Firewall rules documented. Egress filtered by category. All open ports are intentional and documented. Network topology is mapped and owned.

Controlled

Ingress restricted to known, authenticated sources. Egress allow-listed by specific endpoint. Zero-trust network access enforced. All connections logged.

Closed

Circular topology -- nodes communicate only with other known, authenticated nodes in the governed network. Mutual TLS between all peers. No public-facing service endpoints.

Minimal

Single auditable egress point. Every network crossing is logged as a first-class audit event. Certificate-pinned external endpoints only. There is no surface to fingerprint, probe, or chain across.

04 Boundary

How completely do you govern external AI inference endpoints and AI agent access?

Uninspected

Unrestricted external API calls to AI inference providers. No logging, no validation, no endpoint registry. AI agents with computer-use capability may have unmanaged access to the governed boundary.

Inventoried

External AI inference endpoints are documented. Computer-use agent access is identified. No formal controls beyond documentation.

Controlled

Rate limited, logged, response validated. No unmanaged AI computer-use agent access to the governed boundary. Endpoint list maintained and current.

Hardened

Certificate pinned per endpoint. Request signing active. Anomaly detection and circuit breaker deployed. Payload sanitization on all AI inference responses. Governed endpoint registry with change control.

Attested

All hardening controls active. GABA attestation complete for every endpoint. Residual risks formally documented and signed by an authorized signatory. Every boundary crossing logged as a first-class audit event. Computer-use agents explicitly excluded or individually attested.

Minimum Surface Score

Read the CRC Standard Talk to General Reasoning

General Reasoning, Inc. · Birmingham, Alabama · MIT License · 2026
Enterprise inquiries: inquiries@genreason.com