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.
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
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.
Other platform modules
Deploy AI with confidence.
Start with an independent assessment of your highest-stakes AI system.