Meridian GPT — Service, System & Model Cards (Example)

See them as documents. Each card below is also rendered as an actual institutional document — classification-stamped, with a metadata header, numbered sections, and an approval block, ready to print or save as PDF. Use the View as document links, or jump straight to the Service, System, and Model cards.

How these cards connect

A managed platform is the clearest case: one service runs on one system, and that system offers many models. The cards mirror that shape.

Service card Meridian GPT What users choose from the catalog Institution-wide

runs on · 1 : 1

System card MSU AI Gateway One configured deployment — controls & audit Restricted Internal

offers · 1 : many

Model cardGeneral assistant (large)Vendor

Model cardFast assistant (small)Vendor

Model cardVision-capable modelVendor

Model cardInternal model summaryInternal

A different platform would keep the same shape — swap in its own gateway and its own set of models.

Read another way, those same two cards carry the platform's two governance lanes:

Use lane · Service card

Tool & use governance

Who may use Meridian GPT and what they may build on it — eligible users, approved and prohibited uses, and the permitted / conditional / out-of-scope downstream tools.

Platform lane · System card

Platform governance

How the shared service itself is run — the model catalog and routing, controls, logging, change management, and the governance boundary it enforces.

The two lanes meet at the governance boundary: the system card sets the data-class line (up to Internal), and the service card tells the community what they may build within it.

Service card

Layer 4 · Service Institution-wide View as document

Document Title:    Meridian GPT — Service Card
Document Type:     Service Card
Primary Audience:  Students, faculty, staff
Visibility:        Institution-wide
Owner:            Office of the CIO — AI Services
Approver:         AI Governance Committee
Version:          2.1
Review Date:      2026-05-12

What it is

Meridian GPT is the university's secure, multi-model AI workspace. It lets the campus community chat with approved AI models, upload documents for analysis, and use shared prompt templates — without prompts or files being used to train vendor models.

Audience

All MSU account holders

Access

Single sign-on (SSO)

Cost to user

Free / centrally funded

Data allowed

Public & Internal

Approved uses

Drafting and editing, summarizing non-sensitive documents, brainstorming, coding help, and learning support. Faculty may use it to prepare materials and design assignments.

Not approved

Entering Confidential or regulated data (e.g. SSNs, full student records, health data), or relying on outputs for final grades, admissions, or hiring decisions without human review.

Data guidance

Treat inputs as Internal. Do not paste anything classified above Internal under the Data Classification for AI Use standard. Outputs may be inaccurate — verify before use.

Related artifacts

MSU AI Gateway system card · Model cards · Acceptable Use guidance · Teaching & Learning RMF profile

System card — MSU AI Gateway

Layer 5 · System Restricted Internal View as document

Meridian GPT runs on the MSU AI Gateway — the configured institutional deployment that brokers requests to vendor models, enforces data controls, and produces the audit trail governance relies on. This card is internal.

Deployment

Institution-managed gateway, vendor APIs

Identity

SSO + role-based access

Data residency

Contracted no-training; US region

Logging

Access & usage logs, 90-day retention

Architecture

A web client and API broker sit in front of multiple vendor model endpoints. The broker applies authentication, rate limits, content filters, and data-loss-prevention checks before forwarding requests.

Integrations

Campus SSO for identity, the data-classification service for input checks, and the SIEM for security monitoring. An optional LMS link surfaces Meridian GPT inside courses.

Controls

No-training contractual terms, prompt/response logging for abuse review, DLP on uploads, per-role model availability, and an emergency disable switch.

Governance boundary

The gateway is approved for data up to Internal. Confidential and regulated workloads are out of scope and routed to the Secure Research AI Cluster instead.

Model cards

Layer 6 · Model Mixed visibility View as document

Meridian GPT offers several models. Vendor model cards are linked where public; internal summaries capture local context such as which uses each model is approved for.

Model	Best for	Documented limits	Model card
General assistant (large)	Reasoning, long documents, coding	Can be confidently wrong; knowledge cutoff applies	Vendor (public)
Fast assistant (small)	Quick drafts, high-volume tasks	Less reliable on complex reasoning	Vendor (public)
Vision-capable model	Reading charts, images, scanned text	May misread low-quality images	Vendor (public)
MSU internal summary	Local approval & routing notes	Internal only; not for redistribution	Internal summary

Why mixed visibility: vendor model cards are already public and can be linked openly, while the internal model summary — which records local approval decisions and routing logic — stays Restricted Internal.

Supporting evidence

Layer 7 · Evidence Confidential

Behind the published service sits the assurance trail. These documents are confidential and listed here only to show that they exist and connect to the service.

Privacy and security review for the AI Gateway deployment
Accessibility evaluation of the Meridian GPT web client
Vendor due-diligence and contract terms (no-training, data handling)
Risk register entry with mitigations and residual-risk rating
Quarterly usage and abuse-review report