Robots Atlas>ROBOTS ATLAS
Safety

System Card

2023ActivePublished: 18 November 2026Updated: 18 November 2026Published
Key innovation
Extends transparency reporting from a single model to an entire deployed AI system — covering safety evaluations, red-teaming, risks and mitigations beyond the model itself (usage policies, access controls, monitoring).
Category
Safety
Abstraction level
Pattern
Operation level
SystemEvaluation (runtime)Deployment
Use cases
Transparency documentation at model releaseCommunicating safety evaluations and red-teamingAI risk management and governanceCompliance and accountability (e.g., EU AI Act)Informing users about a system's limitations and risksDue diligence when deploying an AI system

How it works

A System Card is a structured report, usually published when a model or system is released. It typically contains: (1) an overview of the model and system, with deployment context; (2) capabilities and limitations; (3) safety evaluations across risk categories, often within formal processes (e.g., OpenAI's Preparedness Framework or Anthropic's Responsible Scaling Policy); (4) external red-teaming and expert engagement (the GPT-4 System Card used more than 50 experts); (5) a description of identified risks; (6) mitigations at the model level (e.g., safety training, RLHF) and at the product and system level (policies, access controls, monitoring); (7) residual risks and the deployment decision. The document is descriptive — it does not modify the system but documents its behaviour and safety-assurance process.

Problem solved

A Model Card documents the characteristics of a single model, but a deployed AI system also includes non-model components — usage policies, access controls, moderation, tools and monitoring — whose behaviour and risks a model card alone does not capture. The System Card addresses the need for transparency and accountability at the whole-system level, including mitigations and residual risks that only emerge after deployment.

Components

Model and system overviewIntroduction

The scope of the document and a description of the model and the system around it (non-model components, deployment context).

Capabilities and limitationsCharacterisation

What the system can and cannot do, including dual-use capabilities.

Safety evaluationsEvaluation

Results of safety testing across risk categories, often within a Preparedness Framework or Responsible Scaling Policy.

External red-teamingAdversarial testing

Engagement of external experts to adversarially test the system before deployment.

Risk assessmentRisk analysis

Identified risks: bio, cyber, persuasion, privacy, disinformation and others.

MitigationsCountermeasures

Mitigations at the model level (safety training, RLHF) and at the system level (policies, access controls, monitoring).

Deployment and residual riskDeployment decision

Risks remaining after mitigations and the rationale for the deployment decision.

Implementation

Implementation pitfalls
Limited, self-reported scopeMedium

System cards are voluntary and published by the provider itself; the scope is sometimes deliberately narrowed (e.g., the GPT-4 System Card explicitly excluded image capabilities and fine-tuning), which limits completeness.

Fix:Treat a system card as one source among several and complement it with independent audits and evaluations.
Lack of a standardised formatMedium

There is no single standard; different providers report different risk categories and metrics, which hampers comparability across systems.

Fix:Map the content onto common frameworks (e.g., Preparedness Framework / RSP categories) when comparing.

Evolution

2019
"Model Cards for Model Reporting" (Mitchell et al., FAT*)
Inflection point

Introduces the model-reporting card — the foundation that the System Card extends to the system level.

2023
GPT-4 System Card (OpenAI)
Inflection point

Popularises the term "system card"; a system-level analysis with non-model mitigations and red-teaming (more than 50 experts).

2024
GPT-4o System Card (OpenAI)

A system card for a multimodal model; evaluations under the Preparedness Framework and third-party dangerous-capability assessments.

2024
OpenAI o1 System Card

Safety evaluations of a reasoning model, external red-teaming, the Preparedness Framework and deliberative alignment.

2024
Anthropic system cards for Claude models

Anthropic publishes system cards for its Claude models alongside Responsible Scaling Policy evaluations and red-teaming.