ACF
acfstandard.io
Developer docs
FR
KPI

Sovereignty Score™

The missing KPI of agentic governance. A 0-100 measure of the decisional sovereignty your organisation actually retains over its production AI agents. Six weighted dimensions, calibrated on the ACF® cards and the EU AI Act obligations.

Tip
Every organisation measures ROI, NPS, CAC. None measures the decisional sovereignty retained over its AI agents — and that is exactly what regulators will start asking for on December 2, 2027. The Sovereignty Score™ is designed to enter the CFO / CISO / DDAO conversation as a number that can be discussed.

Why a new KPI

An organisation deploys an agent that qualifies leads, sets prices, opens credit lines or books flights. The strategic question is not “how much money does it save” — that answer is easy and shifts every quarter. The strategic question is: how much decisional sovereignty have we kept over it?

The Sovereignty Score™ makes that question quantifiable and comparable. The way NPS did for customer satisfaction in 2003, the way eNPS did for employee engagement, the Sovereignty Score™ gives governance committees, boards and regulators a single number to reason about.

The formula

Composite score, 0-100 scale, computed across six weighted dimensions. Each dimension is anchored to an ACF® card and rated 0-100. The final score is the weighted mean:

Sovereignty Score™ = Σ ( weight_i × dimension_i )

where Σ weight_i = 1.00

The six dimensions

ACF-12weight 18%

Decision identifiability

For every consequential decision the agent makes, can you name — instantly — the human individual or governing body who carries responsibility?

ACF-06weight 18%

Override capability

Can a single human override or reverse a specific agent decision in a bounded time (under 1 hour for critical, under 24 hours for non-critical)?

ACF-08weight 18%

Audit traceability

If a regulator asks for the full reasoning behind a specific decision made last Tuesday at 14:12, can you produce a cryptographically signed trace?

ACF-03weight 14%

Threshold control

Are the autonomy thresholds (financial caps, action types, recipient lists) controlled by humans via documented governance, not by the agent itself?

ACF-06weight 18%

Kill switch effectiveness

Is the three-level kill switch tested in production every quarter, with the test result signed into the audit register?

ACF-11weight 14%

Drift visibility

Are model drift, case-distribution drift, and cost drift observable in real time, with alert thresholds owned by a DDAO?

Reading the scores

ScoreBandReading
0–19Critical exposureNo frame. Major AI Act and GDPR risk.
20–39Weak sovereigntyScattered cards. No active DDAO.
40–59Fragile controlFrame under construction. Blind audit recommended.
60–79Controlled sovereigntyN2 target reached. Optimisations possible.
80–100Full sovereigntyACF® reference level — eligible for Level 2/3 certification.

Calculate yours

The /sovereignty-score/calculate page offers an interactive calculator across the six dimensions. It’s a quick estimation tool — for an opposable, DDAO-signed score, use the MCP tool acf.identify-governance-gaps, which also returns the per-dimension rationale and the Ed25519 signature.

The Sovereignty Index

The Sovereignty Index publishes the median scores per industry (banking, insurance, retail, healthcare, public services…) every year, from an anonymised sample of organisations who have scored themselves via the MCP tool. First report scheduled for Q4 2026.