PwC is OpenAI's largest commercial client. McKinsey co-brands with Amazon Web Services. Capgemini is Google Cloud Partner of the Year. It is not a judgement, it is a documented fact. An advisory firm whose business model depends on the tools it recommends cannot tell you what to refuse. This diagnostic assesses what most AI advisors avoid asking.
PwC
OpenAI's largest commercial client worldwide
McKinsey
40% of revenue tech-enabled, co-branding with Amazon Web Services
Capgemini
Google Cloud Partner of the Year, Bleu partner
This diagnostic in 10 questions assesses the AI governance maturity of your organisation on the five dimensions of the Dependency Governance Framework : dependency mapping, non-delegation thresholds, contractual reversibility, fallback plans and geopolitical stress tests.
It requires no registration and no contact details. The result is immediate and tailored to your exposure level.
10 questions · 4 minutes · Immediate result
Progress1 / 10
Question 1 of 10
Dimension 01 · Mapping
Has your organisation documented its AI dependencies layer by layer, from foundation model to network infrastructure?
For each critical AI process: who controls the model? The cloud? The hardware? Is there a tested alternative?
A
Yes, a complete mapping is available and kept up to date.
B
Partially: some processes are documented, others are not.
C
No, we do not have a consolidated view of our AI dependencies.
Question 2 of 10
Dimension 01 · Mapping
If your primary AI model supplier became inaccessible tomorrow, how long before the operational impact becomes critical?
This scenario covers a regulatory restriction as well as a technical failure or a unilateral commercial decision by the supplier.
A
We have a tested continuity plan: under 48 hours to switch over.
B
We could find an alternative, but with no defined timeframe.
C
The impact would be immediate and we have no identified alternative.
Question 3 of 10
Dimension 02 · Non-delegation
Has your organisation formally defined the decisions that cannot be delegated to an artificial intelligence system?
Final credit scoring, medical diagnosis, strategic arbitration, dismissal, proprietary pricing. These thresholds vary by sector but must exist.
A
Yes, a formal list of non-delegation thresholds is documented and approved by the executive committee.
B
There is an implicit understanding, but nothing formally set.
C
This question has not yet been formally asked in our organisation.
Question 4 of 10
Dimension 02 · Non-delegation
When an AI system issues a structural recommendation, is the traceability of the final decision documented and attributable to a named human accountable?
Decisional responsibility is not delegated to an algorithm. In case of litigation or audit, who answers?
A
Yes, every AI-assisted decision is traced with a named human accountable.
B
On some processes only, not systematically.
C
Traceability is not formalised at that level.
Question 5 of 10
Dimension 03 · Reversibility
Do your contracts with AI suppliers include data portability and reversibility clauses with quantified exit costs?
A contract without reversibility clauses is a lock-in contract. These clauses are negotiated before signing, not after two years of use.
A
Yes, our contracts include explicit and quantified portability and exit clauses.
B
Partially: some contracts have them, others do not.
C
Our contracts do not contain structured reversibility clauses.
Question 6 of 10
Dimension 03 · Reversibility
Have you assessed your exposure to the US Cloud Act on the data your AI systems process?
The Cloud Act applies to any legal entity under US law, whatever the physical location of the servers. European hosting does not protect if the controlling entity is US-based.
A
Yes, a documented legal analysis has been produced and presented to the executive committee.
B
The topic has been raised but without a formal analysis.
C
This exposure has not been assessed at our level.
Question 7 of 10
Dimension 04 · Fallback plans
For your critical AI dependencies, have you identified and tested operational European or open-weight alternatives?
Mistral, OVH, Aleph Alpha, open-weight models deployable on private infrastructure. These alternatives must be documented before they are needed.
A
Yes, alternatives are documented and at least partially tested.
B
We know the alternatives but have not formally evaluated them.
C
We have no identified fallback plan.
Question 8 of 10
Dimension 04 · Fallback plans
Have the AI uses that have developed in your teams without a formal decision been inventoried and evaluated?
Shadow AI, that is ChatGPT, Copilot or Gemini used without a framework, often represents 60 to 80% of real enterprise use. It creates invisible dependencies and undocumented data risks.
A
Yes, an inventory of unsanctioned AI use has been produced and decisions have been made.
B
We are aware of the phenomenon but have no formal inventory.
C
We have no visibility on the informal AI use of our teams.
Question 9 of 10
Dimension 05 · Geopolitical stress tests
Has your executive committee assessed the geopolitical scenarios that could affect your AI dependencies over the next 24 months?
AI governance is an act of leadership. Not a technical question.
This diagnostic is produced by VEIA.AI, an independent AI decision governance advisory firm serving boards, executive committees and institutional leaders in France and Switzerland.
VEIA holds no partnership with any software vendor or systems integrator. The advice produced does not fund a parallel commercial relationship. This is the structural condition of usable advice, and what sets VEIA apart from firms whose business model depends on the tools they recommend.
Extend the diagnostic: measure your Decision Survival Time with the AI Reversibility Stress Test, a 72-hour protocol that quantifies operational resilience to the failure of a critical AI supplier.