Independent verification of what AI systems and organizations claim
V1LA turns a claim about AI into a traceable finding: what was said, what the evidence shows, where it contradicts and what remains unproven. It starts from the outside, and goes deeper only when the organization authorizes it.
What do you want V1LA to verify?
You define the target. V1LA does the discovery.
The complexity belongs to V1LA, not to you
You decide what you want verified. V1LA performs the initial discovery by itself: it finds the public claims, policies, interfaces and documentation, extracts the statements that can be tested and looks for evidence for and against each one.
You do not need to prepare data, connect systems or describe your architecture to begin. Internal access is a later, optional step.
How an investigation runs
Four steps, the same for every target.
- 01
Observe
Analyze public evidence and observable behavior from the outside.
- 02
Detect
Find claims, inconsistencies and open questions.
- 03
Contrast
Put findings to the organization and to additional evidence.
- 04
Verify
State what is proven, what is not and what needs review.
Two layers, one evidence trail
Both layers feed the same claims, the same evidence and the same report.
External / Organizational Layer
No access to internal systems required.
- Public claims
- Public policies
- Websites
- Interfaces
- User-facing processes
- Organizational behaviour
- Public documentation
- Observable outcomes
- Third-party dependencies
- Communications
- Inconsistencies
- Evidence gaps
Internal Agentic Layer — with authorization
Scoped to what the organization authorizes.
- AI agents
- Agentic workflows
- Model interactions
- Prompts
- Tool calls
- Retrieval
- Handoffs
- Logs
- Access permissions
- Decision paths
- Human intervention
- Internal policies
- Evidence provenance
- Traceability
- Reproducibility
Authorization is explicit, scoped and revocable.
What you get
A claims register
Every statement under test, with its source and its status.
An Evidence Map
For each claim: what supports it, what contradicts it, what the organization answered and what is locked behind authorization.
Hashed evidence
Each item is timestamped and hashed so its integrity can be checked again later.
Contradictions and gaps
Contradictions shown next to the claim they challenge, and the evidence still missing.
Human review
People approve or escalate each finding. V1LA shows why.
A verifiable report
A report whose every conclusion links back to its evidence.
What V1LA is not
V1LA is independent verification. It can sit next to these categories, but it is none of them.
- AI governance software
- Governance manages your own policies and processes. V1LA verifies, from the outside, what is actually claimed and shown.
- Monitoring or observability
- Monitoring watches your systems in operation. V1LA tests claims against evidence and does not need to be deployed inside them.
- Compliance software
- V1LA does not certify compliance or replace a legal review. It tells you what is supported by evidence and what is not.
- Cybersecurity or model security
- V1LA is not a penetration test, red team or adversarial-attack tool.