Model robustness
When do models become unreliable?
Investigating how changing conditions expose failures that aggregate scores can miss.
Related workOur research connects model behaviour, data quality, and evaluation methods. We start with the conditions under which a system can fail, then ask what evidence would reveal it.
Questions we investigate
Data quality
What does the data leave out?
Examining missing examples, inconsistent records, and the relationship between data and model behaviour.
Related workEvaluation & assurance
What evidence supports a deployment decision?
Connecting test criteria, observed outcomes, and risk records so that decisions can be examined.
Related workLab notes
Original observations and attributed reading from the wider field of AI evaluation.
Reading note
EU AI Act Regulatory Framework: 2026 Official Reference
The European Commission's AI Act policy hub remains a primary official reference for risk-based AI obligations and enterprise governance planning in 2026.
Source: digital-strategy.ec.europa.eu ↗
AI Governance
Reading note
European AI Office: Governance Reference for AI Act Implementation
The European AI Office provides the official EU coordination point for AI Act implementation, general-purpose AI oversight, and governance guidance.
Source: digital-strategy.ec.europa.eu ↗
AI Governance
Reading note
NIST AI Risk Management Framework: 2026 Governance Reference
NIST's AI Risk Management Framework remains a core public reference for trustworthy AI governance, risk measurement, and control documentation.
Source: nist.gov ↗
AI Risk Management
Reading note
NIST Generative AI Profile: Risk Reference for Enterprise AI
NIST's Generative AI Profile is a practical reference for evaluating generative AI risks, model behavior, monitoring needs, and control evidence.
Source: nist.gov ↗
Generative AI Risk