Agent Infrastructure
API, MCP, A2A, workload identity, authorization, data, observability, FinOps and European regulatory mapping.
Reference →VOCE AI Governance Institute · Agent Conformance · v1.0
Know what you can entrust to an agent.
Examine your system against 48 documented controls. The VOCE tool distinguishes stated commitments from available evidence, highlights issues to resolve and prepares a record you can share with leadership and technical teams.
VOCE · 360° Agent Assurance
One system, four assurance layers.
VOCE treats agentic AI as an institutional system, not a software feature. The public architecture connects infrastructure, evidence-based conformance, continuous assurance through ARGOS, and a human-factors layer for cognition and neuropsychology now available.
The cognitive layer is designed to examine attention, cognitive load, memory, decision fatigue, interface effects and human oversight. It is not an automated psychological diagnosis and does not replace qualified clinical or neuropsychological practice.
Agent Conformance
Forty-eight evidence-based controls that distinguish declaration, documentation, testing and proof in operation.
Assessment →ARGOS
AI Evaluation & Continuous Assurance: re-evaluation after material changes in models, permissions, tools, data, suppliers or incidents.
ARGOS →Human Cognitive & Neuropsychology Assurance
Available now: attention, cognitive load, memory, decision fatigue, human-machine interaction and the quality of human oversight.
Open →Three levels of assurance.
Self Assessment
The institution records its own state and evidence references. Useful for gap analysis and procurement preparation.
Reviewed
VOCE specialists examine architecture, policies, contracts, configurations and supplied test evidence.
Observed
VOCE specialists review evidence and observe selected permissions, logs, revocations, approvals or tests in the operating environment.
Self Assessment
The public self-assessment.
Select the highest agent maturity actually deployed. Controls above that level are removed. Critical gaps remain visible without being diluted into an average score.
The public assessment runs locally in your browser. Nothing you enter is transmitted to VOCE.
—Applicable controls
—Critical gaps
—At target evidence
Control catalogue
48 controls. Stable identifiers. Evidence attached.
The catalogue is intentionally inspectable in HTML. A board, auditor, regulator or search engine can see what VOCE means by agent conformance without executing the assessment application.
Normative control language: English. Control IDs and wording remain identical across language versions.
API & tool contracts
MCP governance
A2A & delegation
Identity & workload trust
Authorization & authority
Data, RAG & memory
Observability & audit
Evaluation & assurance
Economic authority & FinOps
European regulatory overlay
VOCE · Conformance
Reviewed and Observed are not self-declared labels.
A Reviewed or Observed result requires an evidence file, reviewer traceability and a dated scope. It remains a VOCE conformance assessment, not an EU or ISO certification.
Discuss a Reviewed or Observed assessment →