Independent AI Governance

What your AI actually does. Not what it's claimed to do.

Most organizations can describe what their AI is supposed to do. Far fewer can show what it actually does, under load, in production, on the cases that matter. Veracity AI closes that gap.

ASSESSMENT // VRC-0427
SUBJECTAI system : production
STATUSPARTIAL : boundary unmapped
CLAIMED BEHAVIOR100%
ACTUAL BEHAVIOR62%
38% Unaccounted gap: the figure a regulator or acquirer asks you to defend.
drag to recompute gap
04
Defensibility tiers
05
TRACE dimensions
10
Diagnostic questions
EU AI Act
Aligned
Classification framework

Every AI system falls into one of four states.

We classify systems on a single axis: how well the organization can defend what the AI is doing. The classification drives the engagement.

TIER 01 : OPERATIONAL
The AI does what it says, and you can prove it.

Documented behavior, evidence of testing, monitored outputs, clear escalation paths. The bar most organizations think they hit. Most do not.

TIER 02 : PARTIAL
It works in some cases, but the boundary is not mapped.

Performance is real but uneven. The system handles known scenarios well, fails silently on edge cases, and no one is sure where the line is.

TIER 03 : DECORATIVE
The AI exists for the appearance of using AI.

It produces outputs, but they do not change decisions. Often a model layered onto a process that already worked without it.

TIER 04 : GAP
It is doing something you are not prepared to defend.

Behavior diverges from policy, claims, or contracts. Sometimes by design, sometimes by drift. This is where governance has to engage first.

Run assessment

Where does your AI governance actually stand?

Answer three questions for an instant classification. The full diagnostic returns a written read of which tier your systems fall into, in about ten minutes.

veracity@assessment ~ %
Can you produce written evidence of production behavior?
Do its outputs change real decisions?
Does behavior match your public claims and contracts?
INDICATIVE CLASSIFICATION
pending
Answer the three questions above for an instant read.
Take the full diagnostic →
Why now

The compliance gap is widening faster than the tooling.

The EU AI Act is moving from text to enforcement, with major deadlines landing between 2026 and 2027. State-level AI disclosure rules are arriving in the US, and procurement teams at large buyers are requiring documentation most vendors can't produce.

None of this needs to be alarming, but it does need to be answered. The work we do is closer to financial audit than software development: structured questions, evidence, written conclusions, and a paper trail you can hand to a regulator, an acquirer, or your board.

How we approach the work →

Worked examples

What we keep finding in production.

Representative patterns, anonymized. Different sectors, the same shape: a gap between what the AI is claimed to do and what it can be shown to do.

HEALTH SYSTEM

The vendor black box.

A regional health system ran a vendor clinical-prediction tool that informed triage, but could not show it performed as the contract claimed on its own patient population. The audit defined exactly where it held and where it quietly failed.

Evidence and Traceability gaps: only vendor benchmarks existed, with no local validation and no traceable path from alert to documented logic.
FINANCIAL SERVICES

The board question.

A bank board asked, in writing, for the organization's AI risk exposure. No one could produce a defensible answer. We built the system inventory and scored it on TRACE.

Authority gap: capability claims were never reviewed by the engineering team.
HR TECHNOLOGY

The compliance illusion.

An HR platform had an AI policy on paper, but model updates shipped without review. The governance record described a system that was no longer running.

Change gap: modifications went live with no documented approval.
Who does this work

Independent. Methodical. On the record.

Veracity AI is an independent practice, led by founder and principal Gary Trautmann. We do not build AI systems, sell models, or carry a stake in any vendor or framework. The independence is the product.

Read the full background →
  • Ph.D. Organizational Development and Leadership, University of Arizona. Dissertation on AI bias mitigation and governance.
  • TRACE Registered methodology. U.S. Copyright Office, TX 9-584-804.
  • ~3 decades Financial services technology, including problem management at enterprise scale.
  • Since 1986 Formal training in AI and expert systems.
Start here

Find out which tier your AI actually falls into.

Ten questions. About ten minutes. A written read of where your systems stand and the two or three things to address first.

Take the diagnostic →