AI model cards and documentation

By Sam Rivera, Founder, SentinelPanda · June 19, 2026 · 1 min read · AI Governance

A model card is the spec sheet for an AI system — what it does, how it was trained, where it fails. It is becoming table stakes for governed AI.

What a model card is

A model card is structured documentation describing an AI model: its intended purpose and users, the data it was trained on, how it performs (including across different groups), its known limitations and failure modes, and how it should and should not be used. It is the spec sheet that makes a model governable.

Why it is becoming expected

The documentation requirements in ISO 42001 and the EU AI Act's technical-documentation obligations for high-risk systems need a concrete form — and the model card is it. It is also what lets oversight, fairness review, and incident response actually happen, because it captures what the model is in one place.

What to include

• Purpose, intended use and users, and out-of-scope uses.
• Training data sources and preparation; performance metrics, including across relevant groups.
• Limitations, known failure modes, and the human-oversight and monitoring in place.

Cover third-party models too

You will not have full internals for a vendor LLM, but you should still document what you know and rely on — the vendor's model card, your use, and your testing. Keep a card per significant model. SentinelPanda links model documentation to each AI system in the inventory.