Verifiable AI Governance
01 / 09

Verifiable AI Governance.
A common layer of evidence for all your AI.

V-PROOF connects governance, agents, code, models, and human oversight with cryptographic evidence that a third party can verify. It does not replace your stack: it transforms what happens within it into something that can be proven later.

From risk identification to proof of control. V-PROOF connects AI systems, risks, policies, assessments, and approvals within a single governance registry, with audit trail capabilities.

Demo · 1:17
Verifiable AI Governance

Govern AI. Prove the decision.

In just over a minute, see how systems and owners are registered, available controls are evaluated, and evidence that can be sealed is linked.

The Thesis
02 / 09

Governance says
what should have happened.
V-PROOF helps show what actually happened.

Governance does not end when a policy is written. It begins when a decision, an autonomous action, a code change, or a model version can be linked to its context, its controls, those responsible, and its supporting evidence.

01

Governance

Risks, policies, controls, responsible parties, and approvals.

→
02

Execution

People's work, agents, code, models and connected systems.

→
03

Evidence

Context, identity, action, supervision, and outcome.

→
04

Verification

Integrity and traceability that can be verified by a third party.

The platform
03 / 09

Different processes.
The same body of evidence.

Each module works on a different object, but they all converge on the same layer of context, responsibility, evidence, and verification.

AI Agent Governance
04 / 09

Every agent has limits.
Every regulated action can leave evidence.

Agents are changing the nature of governance: They no longer just generate content. They can use tools, call APIs, operate systems, access information, and trigger processes.

Identity

Which agent took action?

Identity, source system, or delegated authority associated with the execution.

Authority

What was it allowed to do?

Permissions, policies and limits in force at the time of the action.

Action

What did it do?

Tool, API, data, resource, or process on which it operated.

Supervision

Who intervened?

Approval, review, or human intervention when the process requires it.

Result

What did it cause?

Outcome and context related to the same evidence of implementation.

Verification

Can it be proven later?

Verifiable cryptographic evidence for auditing, governance, and third-party review.

How it fits
05 / 09

It doesn't replace your stack.
It connects it with evidence.

Security, identity, GRC, MLOps, and runtime tools continue to do their job. V-PROOF adds a common layer for reconstructing and subsequently demonstrating the governance context of an action.

Layer 01

Identity & Access

Who can take action, and under what authority?

↓
Layer 02

Policy / Runtime / Security

What is authorized, restricted, blocked, or monitored.

↓
Layer 03

Tools · APIs · Models · Data

Where the actual execution takes place.

↓
V-PROOF

Verifiable Evidence Layer

Action + context + responsible party + policy + result + oversight.

↓
Result

Audit · Governance · Assurance

Evidence prepared for internal or third-party review.

An evidence infrastructure for your current systems. V-PROOF integrates —using APIs, CI/CD, and connectors—links each relevant control to a V-Seal™ record and allows auditors, customers, and regulators to verify it without relying on V-PROOF or accessing your systems.

In the real world
06 / 09

Documents.
Code.
Models.
Agents.

Documents

Word · Portal · Desktop

Drafting, revisions, approvals, and supporting documentation related to the asset.

Code

GitHub · GitLab

Technical history, analysis, checks, and evidence associated with commits and versions.

Models

MLflow

Versions, controls, results, approval, and evidence of the model.

Agents

Actions · Tools · APIs

Implementation context, authority, policy, oversight, and results.

Audits
07 / 09

Three audits.
A single governance framework.

Code, models, and agents are audited where they already reside, in read-only mode. Each audit records its systems, controls, evidence, alerts, and approvals in AI Governance. They can be contracted separately, and any one can be the first.

Code audit · GitHubThe code, in context.
  • Glossary and technical history of each branch.
  • Proportion of human and AI-assisted work.
  • Code checks with automatic verification.
  • V-Seal of the commit, including its glossary and history.
MLOps audit · MLflowThe models the company trains.
  • Reading from MLflow without writing to it.
  • Six deterministic controls, without AI.
  • Approval before moving to production.
  • V-Seal of artifacts, data and results.
Agent audit · six platformsThe agents run by the company.
  • Vertex AI, Copilot Studio, Azure AI Foundry, Amazon Bedrock, Agentforce, and ServiceNow.
  • Eight deterministic controls, without AI.
  • One approval per published configuration.
  • V-Seal of the agent configuration.
What They Have in Common
A recordA model, an agent, and an assistant are entries in the same AI inventory.
A fileEvidence, controls, alerts, and approvals all under a single audit trail.
A sealA document, a commit, a model, and an agent are all sealed the same way.
A verificationThe same public page to check them, without an account.
Risks and ComplianceControl · Review · Evidence

Pre-shipment inspection.
Post-transaction audit.

The V-PROOF chat feature includes a local filter that analyzes messages before sending them to the AI provider and blocks any that contain personal data, confidential information, or other restricted categories. After the exchange, the audit compares the response with your organization’s reference documentation.

01Before Shipping

Local preventive filter.

The filter checks the message before forwarding it to the AI provider. If it blocks the message, the transmission is stopped and the reason is indicated. If the filter cannot complete the check, the transmission is also stopped.

Pre-chat Screening
02After the response

Contextual audit.

The assessment uses the regulatory and internal documents from the reference repository and the audit instructions. The result is made available for human review.

Assessment, not certification
03Work Context

A lead to follow up on.

The conversation and its evaluations are linked to the project, document, and working version. Stamping the deliverable is a separate step: an evaluation alone does not constitute a V-Seal™.

Context and Evidence
Scope of the audit

This applies to chat flows that incorporate the filter. An allowed message can be sent to the configured provider. Adding documents to the reference repository does not, in and of itself, create new blocking rules. Automatic evaluations support human review; they do not certify regulatory compliance.

Integration
08 / 09

Go where the work is already happening.

The integration is designed around the systems, events, and data capture points that the organization chooses to manage. V-PROOF offers APIs, SDKs, webhooks, and specific integrations depending on the deployment.

Interfaces

Portal · Word · Desktop

Spaces for people and teams who need to work with governance built into the process.

Software

GitHub · GitLab

Connected repositories for analysis, technical history, and code evidence.

Models

MLflow

Reading models, versions, controls, and results.

Enterprise

API · SDK · Webhooks

Integration with corporate processes and systems through defined data capture points.

Current Scope

Synchronization, MLOps controls, and certain evaluations run on demand. Current safeguards operate within V-PROOF flows. Gateway, MCP, and extended discovery capabilities are part of the product’s evolution.

V-PROOF
09 / 09

Control the system.
Control the agent.
Demonstrate the action.

Connect systems, agents, people, code, and models with a single layer of governance and verifiable evidence.