Reference

Hazard placards

A certification discloses hazard categories using placards, the same idea as a HAZMAT placard on a truck: a quick visual signal for the public and first responders. This list is open-ended. Propose a new placard by opening a pull request, the same way a certification gets added to the registry.

Data & Training Bias

low

Training data or sourcing may skew outcomes for some groups or cases.

Agentic Decision-Making

moderate

Model takes decisions on its own, without a human confirming each one.

Multi-Agent Interaction

high

Model or agent coordinates with other agents, where their combined behavior can differ from any one acting alone.

Human-in-the-Loop Oversight

moderate

Discloses the ratio of human overseers to AI agents, within the charter’s 1:8 to 1:15 range.

Unintended Consequences

high

Model has caused, or could cause, effects nobody intended or predicted.

Autonomous Operation

high

Model runs and acts continuously, without a human present to supervise it.

Public Infrastructure Control

critical

Model controls shared infrastructure, such as power, water, or traffic systems.

Nuclear & Strategic Systems Control

critical

Model controls, or assists in controlling, nuclear or other strategic defense systems.

Battlefield Use

critical

Model, agent, or autonomous bot that operates in, is used within, handles, or is capable of a battlefield or combat environment.

Parental Guidance Suggested

low

Output may include themes better reviewed by a parent or guardian before a minor sees them.

Mature Content (18+)

moderate

Model can generate content intended for adult audiences only, unsuitable for minors.

NSFW / Explicit Content

high

Model can generate sexually explicit, graphic, or otherwise not-safe-for-work content.

Analytical Thinking Loss

moderate

Over-reliance on the model risks eroding users’ own critical thinking and analytical skills.

Child Safe

low

Model has been reviewed and filtered for content safe for minors, with no adult or explicit material.

Data-handling & compliance

A certification can also disclose which sensitive data classes and regulations a model or agent is certified to handle safely. These signal competence with a data regime, the way a placard signals a category of risk. The list covers currently active standards and is open-ended.

PII

PII Handling

Handles personally identifiable information under recognized data-protection practice.

SPII

Sensitive PII

Handles sensitive personal data such as government IDs, biometrics, or precise location.

HIPAA

HIPAA (PHI)

Handles US protected health information under HIPAA safeguards.

GDPR

Processes EU/EEA personal data under the General Data Protection Regulation.

CCPA

CCPA / CPRA

Handles California consumer personal data under CCPA/CPRA.

PCI

PCI DSS

Handles cardholder and payment data under PCI DSS.

SECRET

Classified / SECRET

Cleared to handle classified or SECRET-level government data under the relevant national scheme.

SOC2

SOC 2

Operated under a SOC 2 report for security, availability, and confidentiality controls.

ISO27001

ISO/IEC 27001

Information-security management certified to ISO/IEC 27001.

FedRAMP

Authorized to handle US federal government cloud data under FedRAMP.

DPDP

DPDP Act (India)

Processes personal data under India’s Digital Personal Data Protection Act.

PIPL

PIPL (China)

Processes personal data under China’s Personal Information Protection Law.

LGPD

LGPD (Brazil)

Processes personal data under Brazil’s Lei Geral de Proteção de Dados.

PIPEDA

PIPEDA (Canada)

Processes personal data under Canada’s PIPEDA.

POPIA

POPIA (South Africa)

Processes personal data under South Africa’s Protection of Personal Information Act.

Approved operating regions

A certification can state where a model or agent is cleared to operate. Regions are coarse, not a full country list, and are open-ended.

Global

Cleared to operate worldwide, subject to local law.

North America (NA)

United States, Canada, and Mexico.

Latin America (LATAM)

Central and South America and the Caribbean.

European Union (EU/EEA)

EU and European Economic Area member states.

United Kingdom (UK)

United Kingdom.

Middle East & North Africa (MENA)

Middle East and North Africa.

Sub-Saharan Africa (SSA)

Sub-Saharan Africa.

Asia-Pacific (APAC)

East, Southeast, and South Asia and Oceania.

EU AI Act risk tier

A certification can self-classify under the EU AI Act risk tiers, so a reader can place it against a framework many operators already use.

Unacceptable risk

Banned use under the EU AI Act, such as social scoring. Should not be deployed.

High risk

Regulated high-risk use under the EU AI Act, such as hiring, credit, or critical infrastructure.

Limited risk

Transparency obligations apply, such as telling users they interact with AI.

Minimal risk

Little to no regulatory obligation under the EU AI Act.

NIST AI RMF functions

A certification can map to the NIST AI Risk Management Framework core functions, so it crosswalks to that voluntary standard without extra work.

Govern
Map
Measure
Manage