AI risk

A risk level you can explain, factor by factor

MERIDVAR classifies each AI system and vendor by applying published, versioned rules to stated facts. The result is not a score. The record shows exactly which facts and which rules produced the level.

  • Low, Moderate, High and Critical
  • Published, versioned rules
  • Unknown is never read as safe
  • Nothing is inferred from traffic
MERIDVAR consoleIllustrative

Risk classification High

Computed by published rules from the completed baseline assessment.

FactorSourceReaches
Personal data can reach itAssessment answerModerate
A linked MCP server has a tool that can deleteMCP inventoryHigh
Use for training is unknownUnknown is never read as safeModerate

This level requires a privacy review, a security review and an approved vendor before approval can be requested.

AI system factors

The level is the highest that any rule reaches

LevelExamples of factors that reach it
CriticalThe organisation has classified the system as a prohibited practice under the EU AI Act. Outputs are used in decisions about individuals, and no person reviews them. A linked agent or MCP server can take consequential actions, and no person approves them.
HighThe organisation has classified the system as high-risk under the EU AI Act. Special category data, credentials or secrets can reach it. A linked MCP server has an unblocked tool that can delete or run commands. A linked agent can run code, move money, change access or send data outside the organisation.
ModeratePersonal, financial or confidential data can reach it. Outputs are used without human review. Access is not limited to named people. A linked agent or MCP server has no completed review or is not in the inventory.
LowA completed assessment in which no factor applies.

Without a completed assessment, there is no classification. The system reads “Not assessed”, never “Low”.

Principles

Four properties a risk officer can rely on

Unknown is not safe

Where “yes” would trigger a factor, “unknown” triggers one at Moderate.

Deterministic

The same facts and the same rules version always give the same result. Each classification stores the facts it read and their hash.

Never edited

Classifying again creates a new record. The earlier one is kept and marked as superseded.

Visibly out of date

When an assessment is reopened or a fact changes, the classification says so and says which facts changed.

Third-party AI risk

Vendors are classified on what they do with your data

A vendor due diligence assessment feeds its own set of rules. The result gates the approval of every AI system that depends on that vendor.

LevelVendor factors
CriticalThe vendor trains on the organisation’s data, and there is no data processing agreement.
HighThe vendor trains on the organisation’s data. There is no data processing agreement. There is no contract.
ModerateAny of those is unknown. No independent report or subprocessor list has been provided. There is no notice before model changes. There are no exit and deletion terms.
Vendor recordIllustrative

Vendor classification High

From a completed vendor due diligence assessment.

FactorAnswerReaches
Trains on the organisation’s dataUnknownModerate
Data processing agreementNoneHigh
Notice before model changesNoneModerate

The result gates the approval of every AI system that depends on this vendor.

Classification overrideIllustrative

Override recorded

Made by an administrator, with a reason.

Computed level
Moderate
Override level
Low
Reason
Recorded with the override, with optional evidence
Still required
Privacy review, because personal data can reach the system

An override can raise what is required but never lower it.

Overrides

An administrator can override, and the record keeps both levels

  • An override needs a reason and can carry evidence.
  • The record keeps the level computed by the rules beside the new level.
  • An override can raise what is required but never lower it. Lowering a system that handles personal data to Low still requires a privacy review.
  • Nothing is inferred from traffic. An observation never makes a system riskier or safer.

The product views on this page are illustrative, with invented example data.

See MERIDVAR refuse a request on your own network

A working session with the founder, on a host you control.

What happens next

  1. You send the formYour request goes straight to the founder.
  2. We reply by emailWe agree a time and what you want to see.
  3. We run a working sessionWe install the appliance with you, route a test request through it and read the evidence it leaves.