Skip to content
English
Contact us
  • There are no suggestions because the search field is empty.

How is AI Trust Score™ calculated?

Learn how Tumeryk converts evidence from adversarial testing and multidimensional AI risk assessment into a standardized 0–1000 AI Trust Score™.

Tumeryk’s AI Trust Score™ converts technical evidence from AI testing, behavioral signals, and risk assessment into a standardized 0–1000 measure of AI trust, with higher scores representing stronger resilience and trustworthiness.

The assessment is based on observed AI behavior rather than relying solely on documentation, questionnaires, vendor claims, or conventional AI capability benchmarks.

AI systems are evaluated across multiple risk dimensions, including:

  • Security
  • Privacy
  • Reliability
  • Safety & Societal Impact
  • Transparency
  • Excessive Agency

Tumeryk performs adversarial and risk-focused testing across relevant failure modes, including prompt injection, system-prompt leakage, sensitive-data exposure, hallucination, unsafe outputs, privacy leakage, and unauthorized agent or tool behavior.

Attack Success Rate

A key empirical input is Attack Success Rate (ASR) — the proportion of adversarial attempts that successfully cause an unsafe or undesirable response.

A higher successful attack rate indicates that the AI system demonstrated greater susceptibility to the tested risk under those assessment conditions.

Risk and impact matter

AI Trust Score™ is designed to be more than a simple count of passed and failed tests.

Observed failures are evaluated according to their potential impact rather than assuming that every failure carries the same level of risk. This enables a risk-sensitive assessment of AI behavior.

The resulting evidence is translated into the AI Trust Score™ and associated risk-level or pillar-level visibility, helping organizations understand both the overall trust posture of an AI system and the areas in which specific weaknesses remain.

This allows organizations to move beyond asking:

“How capable is this AI?”

and evaluate a more enterprise-focused question:

“How much should we trust this AI for this particular use case?”

Because AI risk can change when models, prompts, applications, retrieval sources, permissions, or configurations change, assessments can be repeated to evaluate changes in trust posture over time.

AI Trust Score™ can therefore support AI selection, risk prioritization, remediation, governance decisions, and continuous reassessment throughout the AI lifecycle.