01
Mandatory cover
Five seed business fields plus a technical envelope make every record identifiable and comparable.
Leadership in Enterprise AI Forum
Enterprise AI, grounded in practice.
LEAIF is a collaborative forum for the practical realities of building, buying, deploying and governing enterprise AI.
LEAIF brings together provider, deployer, governance, risk, technology and regulated-industry perspectives to develop practical, interoperable approaches to enterprise AI transparency and accountability. Its work focuses on the implementation gap between policy and operating reality: what enterprises need to know, declare, verify and update when AI crosses organizational boundaries.
AI Facts concept preview · Working draft
Enterprise AI transparency
Know what you’re dealing with.
AI Facts is a proposed concise view of selected information from the fuller AI Passport record—not a separate certification or replacement for the structured record underneath.
The current working draft starts with five seed cover facts—name, developer/author, supplier, version, and execution operator/environment—while additional universal fields remain under working-group review. AI Facts tests how those facts, plus decision-useful context such as purpose, authority and evidence state, can be presented clearly to people.
A concise view of a richer record
AI Facts does not replace the LEAIF disclosure framework. It surfaces selected, decision-useful information while fuller disclosure, supporting evidence and machine-readable information remain available underneath.
AI Facts
Northstar Policy & Procurement Copilot
Developer / Author
Northstar Technologies (fictional)
Supplier
Northstar Technologies (fictional)
Version
Copilot 2.3 · illustrative
Execution Operator & Environment
Enterprise-operated · approved cloud environment
Purpose & Use Boundary
Procurement policy guidance and approved-supplier research; purchasing decisions require human review
Authority / Material Restrictions
Cannot approve spend, execute purchases or bind the company contractually
Lifecycle Status
Illustrative recordDECLARED • EVIDENCE REFERENCE SUPPLIED
Trust states remain distinct from certification.
Trust states remain distinct from certification.
Illustrative fictional example · Proposed presentation under LEAIF review
The enterprise question is not merely “Was AI used?” It is: What am I acquiring, what is inside it, what may it do, who stands behind each declaration, and what changed after adoption?
A framework, not a least-common-denominator label
A small mandatory starting point. Extensible depth when it is needed.
Simple on the surface. Extensible underneath. AI Facts is the concise human-readable view of the LEAIF record—not the boundary of the framework. The fuller record can carry the additional identity, accountability, purpose, limitations, evaluation, oversight, lifecycle, evidence and reference information appropriate to the system and deployment.
The cover is the entry point, not the entire record. Each layer remains attributable, version-specific and machine-readable so downstream organizations can make and maintain their own deployment decisions.
02
Standard pages
Purpose, dependencies, validation, risk, data, deployment and authority use defined page schemas.
03
Composable roll-up
Model, service, system, agent and deployment Passports reference one another without erasing who made each claim.
04
Change-aware
Material model, supplier, data, control or authority changes can trigger review and downstream notification.
Enterprise AI Passport · illustrative
Talent Intelligence Service
Provider Passport · supplier-declared
01 · NameTalent Intelligence Service
02 · Author / developerAI Engineering Team
03 · SupplierSoftware provider
04 · VersionService 4.8.2
05 · Execution operator & environmentSupplier-operated cloud · customer-selected U.S./EU processing region
The working group is reviewing which additional facts are so fundamental that they belong on every cover rather than on a specialized page. Subject type, declarer, schema version, timestamps, upstream references, state and integrity data remain mandatory in the technical envelope.
Initial Passport page catalog
P
Purpose & Use BoundariesFunction, users, context, intended, conditional, unsupported and prohibited uses.
C
Components & DependenciesModels, tools, providers, versions and upstream Passport references.
V
ValidationTesting criteria, datasets, applicable version, results, limits and evidence state.
R
Risk & ComplianceControls, obligations, threats, mitigations and residual risk.
D
DataSources, categories, permissions, access boundaries, retention and deletion.
DP
DeploymentOperator, runtime environment, region, integrations and configuration.
H
Human OversightReview, escalation, intervention, appeals and accountable ownership.
A
Agent Identity & AuthorityPrincipal, tools, actions, data scope, approval, revocation and logging.
CH
Change HistoryVersions, substitutions, permission changes, currentness and notification.
O
Operations & MonitoringTelemetry, drift, incidents, alerts, rollback and service ownership.
Provider and deployer perspectives
The same organization can occupy several roles.
A company may publish software to customers, deploy that software internally, develop its own models and integrate upstream services. The Passport preserves the role and accountability associated with each declaration.
What are we supplying?
The provider identifies the service, its dependencies, its intended and restricted uses, the evidence available and the changes customers need to know about.
- References exact upstream model and service versions
- Adds end-to-end system validation and operating boundaries
- Notifies customers of material dependency changes
How are we using it?
The deployer adds the specific business process, users, data, controls, decision influence, accountable owner and local evaluation appropriate to the deployment.
- Describes actual data and access boundaries
- Records required human review and escalation
- Maintains responsibility for its own use decision
ModelCreator, version, capabilities and model-level limits.
Service / APICommercially supplied AI capability and operating terms.
SystemModels, retrieval, controls, interfaces and workflows.
AgentIdentity, available tools, actions and authority model.
DeploymentEnterprise configuration, purpose, data and accountability.
Runtime attestationTime-limited principal, permissions, session and integrity reference.
Agent profile
Agents add authority to transparency.
A model produces outputs. An agent may select tools, retrieve data and take actions on behalf of a person or organization. The Agent Identity & Authority Page makes that delegated authority visible without replacing the underlying authorization system.
Interactive working concept
See what is inside the service—and know when something underneath it changes.
The v0.3.1 enterprise demo shows provider and deployer views, Passport pages, a supply-chain roll-up, model substitution and an agent permission expansion.
Standards coexistence
Reference existing mechanisms. Do not replace them.
The AI Passport is intended to connect and present information from provenance, identity, authorization, security, model-card, system-card and assurance mechanisms. It is not a competing content-provenance standard, a universal quality score or a certification that a system is safe or compliant.
C2PA or other provenance evidence can be referenced where applicable.
Authorization policies remain enforceable in the systems designed to enforce them.
Declaration, evidence, verification and currentness remain separate trust states.
Organizations represented in discussions and feedback
Informed by enterprise providers, deployers and governance practitioners.
The working draft has been informed by discussions, demonstrations and written feedback involving professionals associated with the organizations below. Individual names are intentionally not displayed.
SCCGForum convening and enterprise AI initiative development
Kastle SystemsEnterprise technology perspective
LinkedInEnterprise AI and platform perspective
SoFi TechnologiesEnterprise AI and governance perspective
H2O.aiAI governance and model-risk perspective
Fiddler AIAI observability and governance perspective
Oracle CloudEnterprise cloud and AI perspective
AWSCloud and enterprise AI perspective
ServiceNowEnterprise workflow and AI perspective
Intel / Intel LabsEnterprise and applied AI perspective
MicrosoftEnterprise governance perspective
Wells FargoFinancial-services AI perspective
Zodia CustodyRegulated-enterprise perspective
HumanaEnterprise AI perspective
Bank of AmericaModel-risk and financial-services perspective represented in prior experience
Participation notice: Company names identify organizations represented in discussions, demonstrations or feedback contributing to the initiative. Their inclusion does not imply corporate sponsorship, approval, adoption, endorsement or authorization to use a company logo.
Working-group decision agenda
What remains open for v0.3 review.
Complete the coverConfirm the five seed fields and nominate the remaining universal fields.
Assign required pagesDecide which pages are mandatory for each subject and participant role.
Define trust statesSeparate declaration, evidence, verification and currentness with controlled values.
Set change triggersDefine which substitutions, configuration changes and authority changes require notification.
Select a pilotTest model substitution, agent permission expansion or both in an authoring workflow.
Current materials
Review the working concept.
Explore the enterprise demo or download the v0.3 strawman. Both use illustrative data and are intended to support working-group discussion, not to represent an actual disclosure by any named organization.