NEW The 2026 State of AI Trust Report is now available. Read the report →

Assurance Platform

A model that passed in March is not the model you have in September

Ongoing observation of deployed AI systems, so degradation, drift, and silent vendor changes surface between formal assessments rather than after an incident.

Capabilities

What this module provides

Performance drift Monitoring of accuracy, calibration, and output distribution against the baseline established at assessment.
Change detection Detection of model, prompt, configuration, and upstream vendor changes — including changes the vendor did not announce.
Control decay Verification that the controls credited at assessment are still operating, not merely still documented.
Threshold alerting Alerts at defined thresholds, escalating to reassessment or suspension of a Trust Mark where warranted.
Long corridor of illuminated infrastructure racks.

Why it matters

Assurance decays quietly

Nothing announces the moment an AI system stops being trustworthy. The training distribution shifts, an upstream provider swaps a model version, a guardrail is disabled during an incident and never restored. Point-in-time assessment cannot catch any of that.

  • Population and covariate drift
  • Silent upstream model changes
  • Guardrail regression
  • Prompt and config divergence
  • Dependency updates
  • Operator workaround creep
Continuous trust monitoring

Standards

Anchored to published criteria

ISO/IEC 42001:2023
Artificial intelligence management system (AIMS) — the certifiable organisational standard for governing AI.
NIST AI RMF 1.0
The Govern, Map, Measure, Manage functions, plus the Generative AI Profile (NIST AI 600-1).
EU AI Act
Regulation (EU) 2024/1689 — the first comprehensive, risk-tiered statutory regime for AI.

Deploy AI with confidence.

Start with an independent assessment of your highest-stakes AI system.