Authority
What may an autonomous agent do, under which conditions, for how long, and within which scope?
SGAEIA · Research program & open reference architecture
Secure Governed Autonomous Edge Intelligence Architecture
SGAEIA investigates how autonomous and distributed AI systems can operate with useful autonomy while authority remains bounded, attributable, auditable, continuously governed, and revocable.
01Problem
From capability to authority
The security problem changes when AI systems move from generating recommendations to taking consequential action. Autonomous agents may hold identities, credentials, tools, permissions, and delegation paths that let them affect external systems. SGAEIA treats that shift as an authority problem, not only a model-safety problem.
Delegation
How can one agent delegate authority to another without losing attribution, constraints, provenance, or accountability?
Distribution
How can governance survive edge–cloud distribution, local action, intermittent connectivity, and multiple administrative boundaries?
Evidence
How can a system prove what authority existed, which policies applied, what actions occurred, and whether governance remained effective?
02Principles
Governance must be architectural
SGAEIA is an open reference architecture and research program built around a simple thesis: useful autonomy should not depend on implicit trust or static approval. Authority must be explicit, constrained, continuously evaluated, and capable of being withdrawn.
Bounded AuthorityUseful autonomy requires explicit limits on scope, duration, permissions, and consequential action.
Authenticated DelegationDelegated authority remains attributable, constrained, traceable, and connected to its source.
Zero TrustConsequential interactions are explicitly verified rather than implicitly trusted.
Revocation by DesignAuthority can be withdrawn as deliberately as it is granted, including after conditions change.
Continuous GRCGovernance, risk, compliance, and assurance remain active during operation rather than only at audit time.
Evidence-as-CodeGoverned operation produces machine-verifiable evidence suitable for continuous assurance and independent review.
03Properties
Security & governance properties
The public SGAEIA model is expressed through properties and invariants rather than implementation-specific protocols. The objective is to make governance testable without tying the research to one agent framework, model vendor, or deployment stack.
Explicit authority boundaries
No autonomous component should acquire open-ended authority merely because it can perform a task.
Traceable chains of authority
Authority should remain attributable across delegation, collaboration, and multi-agent coordination.
Security-preserving substitution
Replacing an agent, model, or implementation component should not silently weaken the governing security properties.
Continuous verifiability
The system should continuously produce evidence that governance controls remain effective during operation.
04Agenda
Research agenda
SGAEIA provides a common frame for a broader research agenda on governed autonomous intelligence.
Governed multi-agent coordination
How should authority be represented and constrained when autonomous agents coordinate, negotiate, or form temporary collectives?
Governed edge autonomy
How should governance behave when decisions must occur locally, with latency constraints or intermittent connectivity to centralized services?
Continuous assurance
How can policy enforcement and evidence generation support continuous GRC instead of periodic compliance snapshots?
Verifiable delegation
How can delegation remain authenticated, bounded, revocable, and reconstructable across complex chains of autonomous action?
05Research Series
SGAEIA Research Series
The research series develops the architecture incrementally through problems, principles, security properties, governance concepts, and emerging risks. Archival versions are preserved on Zenodo and public essays are published on Medium.
2026-09-22
Zero Trust for Multi-Agent AI Systems
SGAEIA Research Series — Article 3
2026-09-20
Ten Days That Exposed the Governance Gap in Frontier AI
SGAEIA Research Series — Article 12
2026-09-18
From Model Capability to Governed Action: An Architecture for Secure Agentic AI
SGAEIA Research Series — Article 11
2026-09-16
Rogue AI Agents: When Autonomous Systems Become a Collective
SGAEIA Research Series — Article 10
2026-09-14
When AI Begins to Build AI
SGAEIA Research Series — Article 9
2026-09-17
Security-by-Design for Distributed Agentic AI
SGAEIA Research Series — Article 7
2026
Evidence-as-Code: Making AI Governance Verifiable
SGAEIA Research Series — Article 6
2026
Continuous GRC for Agentic AI
SGAEIA Research Series — Article 5
2026
Authenticated Delegation Between Autonomous AI Agents
SGAEIA Research Series — Article 4
2026
Why Autonomous AI Agents Need Bounded and Revocable Authority
SGAEIA Research Series — Article 2
2026
From Edge AI to Governed Autonomous Edge Intelligence
SGAEIA Research Series — Article 1
06Artifacts
Research artifacts & persistent records
SGAEIA is published as a versioned research software artifact and supported by persistent archival records, a dedicated Zenodo community, and public research writing.
SGAEIA software artifact
Versioned research software artifact and public reference architecture.
SGAEIA Zenodo Community
Archival collection for SGAEIA research outputs and related publications.
SGAEIA GitHub repository
Public repository for the software artifact and research-facing project materials.
SGAEIA Research Series on Medium
Public essays and research notes developing the SGAEIA research program.
SGAEIA on LinkedIn
Institutional page for SGAEIA research updates, publication announcements, and project communication.
07Citation
Citation
For the SGAEIA software artifact, use the following citation:
Silva, Aridio. (2026). SGAEIA: Secure Governed Autonomous Edge Intelligence Architecture (Version 0.3.4). Zenodo. https://doi.org/10.5281/zenodo.22557796Software license: Apache-2.0. Research articles are archived separately under their published record licenses.