01
A claim is not a behavior
A stated control describes what should happen. A buyer may need to understand how the system actually behaves.
AI ASSURANCE · IN DEVELOPMENT
AI trust takes more than a promise.
We are building an independent AI assurance service focused on observable behavior and evidence that is useful to both AI vendors and enterprise buyers.
We are at an early stage, speaking with teams bringing AI products into enterprise environments.
01 / THE CHALLENGE
Claims matter.Evidence changes the conversation.
Policies, documentation and internal testing are essential. When an AI system enters an enterprise workflow, though, additional questions may call for further assessment.
02
Evidence does not last forever
When a system changes materially, it matters whether earlier findings still apply.
03
Trust needs a shared language
Engineering, security and procurement teams view risk differently. They need precise findings with a clear scope and explicit limitations.
02 / OUR APPROACH
A new layer of clarity.
We do not want to replace internal controls. We want to make their outcomes easier to assess and communicate when decisions matter.
Look at the real system.
Focus on observable AI behavior in an agreed scope, not solely on statements in documents.
Make evidence useful.
Findings should be understandable in context, with the assessment scope and its limitations made explicit.
01 / CLARITYPurposeful assessment
Evaluation suited to the system’s intended context.
02 / RIGORProportionate conclusions
Claims should never exceed the supporting evidence.
03 / CONTINUITYRevisitable evidence
A new system version raises new questions about earlier findings.
03 / HOW IT WORKS
From a real system to a better-informed decision.
The client journey should be straightforward. The full technical methodology is defined with the team, within an agreed scope.
01 / Shared scope
Understand the context
Start with the system, its intended use and the questions that matter to the team and its customers.
02 / Contextual findings
Observe behavior
Conduct a focused assessment in agreed conditions, without assuming every AI response is deterministic.
03 / Explicit limitations
Make the outcome clear
Organize observations into understandable evidence and findings, separating what was assessed from what remains uncertain.
04 / Over time
Know when to revisit
When a system changes materially, existing evidence may need to be updated. That is the principle behind our continuous assurance vision.
04 / WHO IT’S FOR
For teams bringing AI into the enterprise.
We are exploring where greater clarity about AI behavior can help vendors and buyers make better-informed decisions.
01 / 04
Legal AI
Output reliability, appropriate source use and confidentiality of sensitive information.
02 / 04
Financial AI
Controls, operational boundaries and interactions with sensitive data or processes.
03 / 04
HR & Recruiting AI
Candidate data handling, consistent behavior and human oversight.
04 / 04
AI Agents
Autonomy, access boundaries and behavior across tools and enterprise workflows.
THE GOAL
Evidence matters when you know what it applies to.
Our vision is to provide an independent, understandable reference on AI-system behavior without mistaking an assessment for an absolute guarantee of safety.
The full platform is being developed. Early partnerships will also help establish its value and appropriate scope.
05 / DESIGN PARTNERS
Build trust together.
If you sell an AI product to enterprise customers, we would like to understand what evidence they ask for, how you prepare it today and what would make it genuinely useful.