AWS MSP PROGRAM // VCL 8.0

AWS MSP VCL 8.0: measurable evidence for the AI controls, not just a policy document

From 1 January 2027 AWS validates MSP partners exclusively against Validation Checklist 8.0: 61 controls, 24 new, 27 rewritten, focused on demonstrable AI delivery. monsys measures the technical controls — GenAI observability, toil reduction, AI dependencies in vulnerability management, signed change management — and bundles them in an Ed25519-signed evidence pack an AWS auditor can verify independently.

61
controls in VCL 8.0, of which 24 new and 27 rewritten
2026-12-31
VCL 7.1 sunsets; from 2027 only 8.0 is accepted
15
controls monsys supplies measurable evidence for today
VIDEO // 2:18

See it in two minutes

From the control mapping on this page to the dashboard, the signed evidence pack and the docs — with realistic demo data (Northwind Cloud). English subtitles.

Demo tenant with synthetic data. No customer data.

What changes in VCL 8.0

The previous checklist asked whether you could deliver AI. The new one demands proof that you operate it: observability, cost, governance and toil reduction have to come out of systems, not slides.

GenAI and agentic observability (OPS-004)

New. Agent session tracing, inference performance monitoring, drift detection, decision audit trails and cost observability. Every MSP must show this for its own AI workloads and those of its customers.

Toil as a metric (OPSP-003)

New. Toil must be tracked as a percentage of ops time, with an automation backlog and demonstrated quarterly reduction. That needs a time series, not a snapshot.

AI inside existing controls

Rewritten. Change management (GOVP-001), vulnerability management (SEC-003) and FinOps (OPS-009) now explicitly cover AI workloads, AI dependencies and inference cost.

Control by control: what AWS asks, what monsys measures

Not a marketing list but the actual mapping shipped in the platform (compliance_framework_controls, framework AWS-MSP-VCL-8.0). Every row has an executable evidence query; the dashboard shows per control how much evidence exists.

●automatic: a monsys signal proves it◐partial: signal + MSP attestation○declarative: attestation in the evidence pack
ControlWhat VCL 8.0 asksWhat monsys measuresCoverage
OPS-004
GenAI & agentic AI observability
Agent session tracing, inference performance monitoring, drift detection, decision audit trails and cost observability.AI traces and spans per app (tokens, latency, model, refusal, PII hits), agent sessions per session hash, drift and cost rules as alerts.●automatic: a monsys signal proves it
OPS-009
FinOps incl. AI cost
Track and optimise inference cost per workload, GPU right-sizing.Cost per trace/app/model in euro with pricing snapshot; cost-anomaly alerts per AI app.●automatic: a monsys signal proves it
OPSP-003
Toil measurement and reduction
Toil as a percentage of ops time, an automation backlog, demonstrated quarterly reduction.Monthly automation ratio (auto vs manual actions), auto-resolved alerts, MTTA/MTTR, quarter delta — per tenant, never per person.●automatic: a monsys signal proves it
SEC-003
Vulnerability management incl. AI deps
Track vulnerabilities in AI model and agent dependencies.OSV/EPSS/KEV on app dependencies (pypi, npm, Go), OS packages, kernel and container images; fix availability per CVE.●automatic: a monsys signal proves it
OPS-010
Patch and release management
Documented, repeatable patch execution.Signed patch actions with result and latency per host; auto-patch for security updates.●automatic: a monsys signal proves it
GOV-003
Data governance for AI
Protection of sensitive data in AI workloads.PII redaction at source, redaction level per AI app; no raw prompts reach the hub.●automatic: a monsys signal proves it
GOVP-001
Change management incl. AI workloads
Approved, traceable changes to infrastructure, applications and AI workloads.Emergency Action Tokens: Ed25519-signed, single-use, short TTL, with result and optional multi-approval.◐partial: signal + MSP attestation
SEC-007
Agentic zero trust & blast radius
Containment of what an (AI) agent can reach.Per-host exposure (centrality, internet-facing), tokens bound to one host and one action list.◐partial: signal + MSP attestation
SECP-001
Encryption and key management
Encryption in transit/at rest, key rotation.mTLS hub↔agent with per-agent certificates, signed evidence packs; key management of your AWS workloads is attested by you.◐partial: signal + MSP attestation
SEC-001
Identity & Access Management
Least privilege, MFA, periodic access review.Scoped roles, TOTP on admin access, quarterly access-review report for the monsys platform.◐partial: signal + MSP attestation
SECP-004
Prompt injection prevention
Input validation, content filtering and guardrails on AI endpoints.Refusal and anomaly detection on traces; the guardrails themselves live in your app.◐partial: signal + MSP attestation
GOV-002
AI agent and model governance
Inventory of agents/models with owner, risk tier and lifecycle.AI app register with owner, risk tier and allowed models; undeclared models surface in traces.◐partial: signal + MSP attestation
PLAT-004
Agentic AI platform
Safe platform for agentic workloads with human control.Human approval before every action (token + TOTP); never autonomous LLM execution.◐partial: signal + MSP attestation
PLAT-005
Well-Architected
Continuous architecture review.Trust Score per tenant as composite health/risk metric; the AWS WAR itself is done with AWS.◐partial: signal + MSP attestation
GOVP-002 et al.
Responsible AI, roadmap, practice lead, FDE
Policy, AI transformation roadmap, vertical specialisation, people.Attestation checklist with signature line in the evidence pack; monsys supplies the measurements the roadmap steers on.○declarative: attestation in the evidence pack

monsys supplies the measurements per control; the AWS validation partner judges the file. The dashboard shows the evidence count per control — including where it is zero.

Measure toil without measuring people

OPSP-003 asks for toil as a percentage of ops time. Nobody measures human minutes reliably, but the proxy auditors accept is measurable: what share of remediation actions on your fleet was automated, and how many alerts resolved themselves.

  • Per month: total executed actions, automated vs manual, automation ratio in %.
  • Auto-resolved alerts, MTTA and MTTR from the same time series.
  • Quarter delta: mean of the last three months against the three before. Without two full windows monsys shows 'n/a' — a trend is never invented.
  • API and report: GET /api/v1/ops/toil and section 3 of the evidence pack.
WORKS COUNCIL // BETRIEBSRAT

Per tenant, never per person

In Germany, Belgium and the Netherlands, per-employee productivity measurement can require works-council consent. That is why monsys aggregates toil per tenant per month and agent sessions per hashed session — there is no table of per-operator performance.

So you can show the numbers to AWS and to your works council without a separate procedure. FINMA-bound (CH) and Baltic MSPs with strict data-residency requirements keep everything on their own EU infrastructure.

The VCL 8.0 evidence pack

One signed PDF + JSONL bundle, generated from the report centre, offline-verifiable with your tenant's public key. Six sections:

  1. Control coverage: all mapped controls with level, evidence count and review status.
  2. GenAI & agentic observability per AI app: traces, tokens, cost, refusals, tool calls; agent-session KPIs.
  3. Toil & automation: six months, quarter delta, method.
  4. Vulnerability management incl. AI dependencies: per ecosystem and severity, KEV-listed, fix available.
  5. Change management & blast radius: signed actions auto/manual, failure rate, highest-exposure hosts.
  6. Attestation checklist for the declarative controls, with a signature line for the MSP.

How to set it up

  1. Install the agent on your managed Linux/Windows hosts (one command). Patch, CVE and change evidence accumulates automatically from then on.
  2. Register your AI apps and ship traces via the SDK with a session hash. PII is redacted at source.
  3. Open Security & compliance → AWS MSP VCL 8.0 in the dashboard: per control you see how much evidence exists and what is still missing.
  4. Generate the evidence pack monthly and add the attestation. Ready for your AWS validation file.

What monsys does not do for you

No compliance verdict

monsys supplies measurements and signed evidence. The AWS validation partner judges the file; we do not replace that judgement.

No guardrails inside your AI app

SECP-004 requires input validation on the endpoint itself. monsys detects the consequences (refusals, anomalies, cost spikes) but does not build the guardrail.

No roadmap, practice lead or FDE team

BUS-002, PEO-003 and PEO-004 are organisational controls. You attest them yourself; monsys supplies the measurements the roadmap steers on.

No replacement for AWS-native tooling

CloudWatch, Security Hub and Cost Explorer stay. monsys adds the host and AI layer AWS does not see, and signs the evidence.

Frequently asked questions

When do I need to be ready for VCL 8.0?

VCL 7.1 sunsets on 31 December 2026. Every validation or revalidation from 1 January 2027 runs on 8.0. Because OPSP-003 asks for a quarterly trend, you need at least six months of time series before your audit — so start measuring in 2026.

Does monsys cover all 61 controls?

No. monsys supplies measurable evidence for 15 technical controls and an attestation checklist for the declarative ones. Business, people and policy controls remain your work. We show honestly per control what is automatic, partial or declarative.

What if I have no AI workloads yet?

Then OPS-004 and OPS-009 show 'no evidence' and you see that immediately on the dashboard. The toil, CVE and change controls work from day one on your ordinary servers. Register your first AI app as soon as it exists; the SDK is open source.

Does this also work for my customers' AI workloads?

Yes. Every customer is a separate tenant with its own AI apps, traces and evidence pack. As an MSP you see everything in one dashboard and generate a signed pack per customer.

Where is the data?

On servers in Belgium, operated by GoTrust BV. No US cloud in the path for storage, alerts or authentication — relevant for FINMA, NIS2 and data-residency requirements.

What does it cost?

5 servers are free, then €3 per server per month. AI observability is part of the platform; the VCL 8.0 evidence pack is a standard report template at no extra cost.

Start measuring in 2026, validate in 2027 with evidence

Five servers free, EU-hosted, no credit card.

Read on
How-to: a VCL 8.0 gap analysis in one afternoon →monsys for MSPs →AI observability →Documentation ↗
Guides

Prefer to do it yourself? Read the guides.

Guides that solve everything by hand first, and only then show what monsys automates.

All guides →

WHITEPAPER

The full story in 25 pages

Architecture, Trust Score formula, supply-chain pipeline, NIS2 / AI Act / CRA mapping. Free PDF after a short form.

Download whitepaper →