Declaration Layers and the Evaluation of Agent Boundaries

APA

Capucci, H. A. (2026). Declaration Layers and the Evaluation of Agent Boundaries. https://doi.org/10.5281/zenodo.18880520

BibTeX

@misc{amw-014,
  title = {Declaration Layers and the Evaluation of Agent Boundaries},
  author = {Capucci, Hernán Alfredo},
  year = {2026},
  month = {3},
  doi = {10.5281/zenodo.18880520},
  url = {https://agent-manifest-spec.org/works/declaration-layers},
  language = {en},
  keywords = {agent-boundaries, pre-execution-authority, declarative-boundaries, agent-manifest, agent-specification, ai-specification, agent-governance, ai-standards, llm-agents, autonomous-systems, json-schema, ai-safety},
  note = {Working paper. Concept DOI: 10.5281/zenodo.18880519. Version: v1}
}

RIS

TY  - GEN
TI  - Declaration Layers and the Evaluation of Agent Boundaries
AU  - Capucci, Hernán Alfredo
PY  - 2026
DA  - 2026/03/05
DO  - 10.5281/zenodo.18880520
UR  - https://agent-manifest-spec.org/works/declaration-layers
LA  - en
AB  - This paper examines the role of declaration layers in alignment evaluation for autonomous AI systems. Building on recent research on scheming and deceptive behaviors in AI systems, it proposes that explicit boundary declarations may complement behavioral alignment approaches by reducing structural ambiguity in agent evaluation contexts. The paper introduces the concept of declaration layers as a governance-oriented interface through which AI agents disclose operational authority, constraints, and accountability structures prior to execution. Agent Manifest is presented as an example of such a declarative infrastructure. By clarifying agent boundaries before action, declaration layers may improve interpretability of behavioral evaluations and support institutional oversight frameworks for increasingly autonomous AI systems.
KW  - agent-boundaries
KW  - pre-execution-authority
KW  - declarative-boundaries
KW  - agent-manifest
KW  - agent-specification
KW  - ai-specification
KW  - agent-governance
KW  - ai-standards
KW  - llm-agents
KW  - autonomous-systems
KW  - json-schema
KW  - ai-safety
N1  - Working paper. Concept DOI: 10.5281/zenodo.18880519. Version: v1
ER  -