System card

A system card extends the model card concept to cover an entire AI system — the model plus prompts, tools, retrieval, guardrails, and intended deployment context.

Where a model card documents a single model, a system card documents the whole AI feature: which model is used at which step, what tools it can call, what RAG sources feed it, what guardrails sit around it, and what use cases are in/out of scope. Anthropic, OpenAI, and Google all publish system cards for major launches (e.g. ChatGPT, Claude.ai, Gemini AI Overviews). For internal production teams, maintaining a one-page system card per AI feature is best practice for safety review, on-call debugging, and audit. It's also the document you hand to a security or legal reviewer.

When to use system card

• Any production AI feature.
• Safety / legal / compliance reviews.
• Operations runbooks for AI features.

Common mistakes

• Conflating with the model card — system cards cover the whole pipeline.
• Letting the system card drift out of sync with the deployed system.

FAQ

Related terms

• Model card — A model card is structured documentation accompanying a released model — what it does, what it was trained on, its evaluation results, intended uses, and known limitations.
• Evals (LLM evaluations) — Evals are systematic tests that measure how well a language model or LLM-powered system performs on a defined task using a golden set of inputs and reference outputs.
• Guardrails — Guardrails are deterministic checks layered around a language model to prevent unsafe, off-topic, or non-compliant outputs from reaching the user.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/system-card.md.