EM-AI-01 · Subject model · W2
AI system, agent and execution
AI system in its context of use, agent configuration, tools, execution and endpoint. The subject of authority remains a defined operator, not automatically a new person.
Partially published; research continues
Claude: complete-frozen-study-and-audit; Grok: complete-browser-study.
Published results
Research note, in Russian: Canonical promotion remains held by the limits published with the profile.
Subject boundary and candidate types
- AISystem
- AgentConfiguration
- ToolBinding
- AgentRun
- InferenceEndpoint
Deep research questions
- How does a system differ from a model and a product?
- How are agent actions and delegation restricted?
- How to reproduce an execution when the model and tools change?
Verifiable invariants
- The configuration and the run have different IDs
- Delegation is limited to the original rights
- An ai-subject is not created as a human or legal subject
End-to-end acceptance scenario
One agent uses two models and three tools: each run records the configuration, the permissions and the verifiable actions.
Negative case
An LLM model receives unlimited authority solely because of the field role=agent.
Approaches to compare
- NIST AI RMF: context, evaluation and risk management
- PROV/SPDX: provenance of data, artifacts and actions
- Model registry/evaluation tooling practice; map system, artifact, run and endpoint
Candidates in the live catalogue
- WM-AI-001 · AI System · 0.3.0-research.1 · installable
Semantic fit requires boundary research; a published model does not by itself complete this card. - WM-AI-002 · AI Agent · 0.3.0-research.1 · installable
Semantic fit requires boundary research; a published model does not by itself complete this card. - WM-AI-005 · Prompt / Agent Configuration · 0.3.0-research.1 · installable
Semantic fit requires boundary research; a published model does not by itself complete this card. - WM-AI-004 · AI Inference / Agent Run · 0.3.0-research.1 · installable
Semantic fit requires boundary research; a published model does not by itself complete this card.
Result requirements
Every card is executed together with the full research contract: definitions, fields and cardinalities, lifecycle, sources, data mastership, rights, the five object facets, at least eight invariants, positive and negative examples, dependencies, migration and applicability limits.