AI Model Transparency Marketing Claim Verification

Understanding the importance of this work, UL Solutions incorporates 11 safety principles into our AI verification. Our transparency benchmarking program spans the technical, ethical and governance pillars, because they collectively address how AI systems function (technical pillar), the fairness and integrity of their decisions (ethical pillar) and the accountability structures necessary to manage and mitigate risks (governance pillar).

Using the UL Solutions AI Model Transparency Benchmark, we verify marketing claims so that manufacturers can offer substantiated statements about areas such as model development, security and more.

AI model transparency is the way we understand how AI systems make decisions, why they produce specific results and what data they are using. A transparency assessment determines how well a software builder documents and communicates their design intent.  

To examine key areas such as data management, model development, security, deployment and ethical considerations, the benchmark provides an objective rating of an AI model’s transparency and trustworthiness.  

Our experts perform science-based assessments and review documentation for best practices as we analyze a company’s AI statements.

We conduct our testing based on our 11 safety principles, using a 100-point scoring scheme to determine the degree of model transparency. This assessment primarily focuses on AI dataset disclosure and description, the overall architecture of the AI system and user consent description.

If a claim meets these requirements, it can display a UL Verified Mark for AI Model Transparency, and — depending on its score — the recipient can receive a rating from Silver to Diamond for achieving a score above 50.