AWS MSP VCL 8.0

The AWS MSP Program Validation Checklist (VCL) 8.0 becomes the only accepted checklist for AWS MSP Partner (re)validation on 1 January 2027; VCL 7.1 sunsets on 31 December 2026. Version 8.0 has 61 controls, 24 new and 27 rewritten, and shifts the emphasis to demonstrable AI delivery: observability of GenAI/agentic workloads, toil measurement, AI governance and AI dependencies in vulnerability management.

monsys measures the technical controls on that list and bundles the evidence in an Ed25519-signed evidence pack. The declarative controls (policy, roadmap, practice lead) are attested by the MSP on the last page of that pack.

Measured by monsys, judged by the auditor

monsys supplies per control the measurement and the signed evidence; the AWS validation partner judges the file. The control → monsys-signal mapping lives in compliance_framework_controls (framework AWS-MSP-VCL-8.0) and is active.

Dashboard page

Sidebar → Security & compliance → AWS MSP VCL 8.0 (/compliance/VCL8). Above the generic control table sit three KPI blocks:

Block	Controls	Source
GenAI & agentic AI observability	OPS-004, OPS-009, GOV-002, GOV-003	ai_apps, ai_traces, ai_agent_sessions
Toil measurement & reduction	OPSP-003	toil_metrics
Vulnerability, change & access management	SEC-003, GOVP-001, SEC-001, SECP-001	inventory_dependency_cves, os_package_cves, emergency_tokens, users, agent_certificates

Below that is the control table as for NIS2/CRA/ISO 27001: id, description (NL/FR/EN), coverage level, evidence count and review status.

What monsys measures per control

Control	What VCL 8.0 asks	monsys evidence	Level
OPS-004 GenAI & agentic AI observability	agent session tracing, inference performance, drift, decision audit trail, cost	ai_traces + ai_spans (tokens, latency, model, refusal, PII hits), ai_agent_sessions (per session hash), ai_alerts (drift/cost rules)	automatic
OPS-009 FinOps incl. AI cost	inference cost per app/model	total_cost_micro_eur per trace, pricing snapshot	automatic
OPSP-003 Toil measurement	toil as % of ops time, quarterly reduction	toil_metrics: automation ratio, auto-resolved alerts, MTTA/MTTR per month + quarter delta	automatic
SEC-003 Vulnerability mgmt incl. AI deps	scan AI model/agent dependencies	OSV/EPSS/KEV on app dependencies (pypi/npm/Go), OS packages, kernel, images	automatic
OPS-010 Patch & release	documented patch execution	remediation_latency per host, signed result	automatic
GOV-003 Data governance for AI	PII protection	redaction level per AI app, PII redaction at source	automatic
GOVP-001 Change management incl. AI workloads	approved, traceable changes	Emergency Action Tokens: signed, single-use, with result	partial
SEC-007 Agentic zero trust & blast radius	containment of agent actions	per-host exposure (centrality, internet-facing), single-use tokens with action list	partial
SECP-001 Encryption & key management		mTLS with per-agent certificates, Ed25519-signed evidence	partial
SEC-001 IAM		scoped roles, TOTP, access-review report	partial
SECP-004 Prompt injection prevention	guardrails on AI endpoints	refusal and anomaly detection via ai_alerts; guardrails themselves are app-side	partial
GOV-002 AI agent & model governance	inventory with owner and risk tier	ai_apps with owner_email, declared_risk_tier, declared_models	partial
PLAT-004 Agentic AI platform		human approval before every action (EAT + TOTP), never autonomous LLM execution	partial
PLAT-005 Well-Architected		Trust Score per tenant	partial
GOVP-002, BUS-002, BUS-006, PEO-003, PEO-004	policy, roadmap, specialisation, people	attestation checklist in the evidence pack	manual

Toil metric (OPSP-003)

ToilMetricsWorker recomputes daily, per tenant per calendar month:

• total_actions — executed Emergency Action Tokens
• automated_actions — tokens issued by a monsys worker (reason starts with auto-: auto-patch, auto-update-all)
• manual_actions — tokens issued by a person
• automation_pct — automated / total × 100
• total_alerts, auto_resolved_alerts (resolution_notes LIKE 'auto:%')
• mtta_seconds, mttr_seconds

The quarter delta compares the mean automation_pct of the last three months with the three before. Positive = more automation = less toil. Without two full windows the hub shows n/a — a trend is never invented.

Works council / Betriebsrat

The metric is deliberately per tenant per month, never per operator. No per-person productivity is stored, so the figures can be shared in DE/BE contexts without a consent procedure.

API: GET /api/v1/ops/toil?months=6

Agent sessions (OPS-004)

AIAgentSessionsWorker rolls up the last 90 days of ai_traces hourly per (tenant, app, user_session_hash): number of traces and spans, tool calls, tokens in/out, cost, models used, refusals, errors, PII hits. The hash is set by the SDK; the hub never sees a raw session id or user identity.

API: GET /api/v1/ai/agent-sessions?days=30&app_id=&limit=100

Evidence pack

Reports → AWS MSP VCL 8.0 evidence pack (aws_msp_vcl8, tenant scope, admin). Sections:

1. Control coverage (all mapped controls, level, evidence count, review status)
2. GenAI & agentic observability per app + session KPIs
3. Toil & automation (6 months, quarter delta, method)
4. Vulnerability management incl. AI dependencies (per ecosystem/severity, KEV, fix available)
5. Change management & blast radius (signed actions auto/manual, failure rate, top-exposure hosts)
6. Attestation checklist for the declarative controls, with signature line

The pack is Ed25519-signed and offline-verifiable like every other audit pack (monsys-verify). Toil and session aggregates are recomputed fresh before rendering.

What monsys does not do

• No compliance verdict: monsys supplies measurements, the auditor judges.
• No guardrails inside your AI app: SECP-004 requires input validation on the endpoint itself; monsys detects the consequences (refusals, anomalies).
• No measurement of human hours: toil is a proxy based on automated vs manual actions.
• No roadmap, practice lead or vertical specialisation: you attest those yourself.

Related

• Compliance coverage
• Reports
• Audit pack
• Emergency Action Tokens