TRUST Framework

The TRUST Framework is Global Data innovation’s proprietary solution for AI governance and data leadership. It builds business velocity, resilience, cost-savings, efficiencies, and minimizes risks by distilling what is key from the legal frameworks  to avoid negative headlines.  The approach is informed by regulators, data scientists, academic trends and the real-life lessons learned from hundreds of AI case studies across sectors.

Triage: Classify AI use cases by risk like a traffic light. Red light for prohibited AI, yellow for high risk and green for medium/low. High-risk applications, such as those in health care and criminal justice, will need to undergo especially rigorous testing and oversight. Integrated into this triage is a focus on melding AI use cases with enterprise goals, and ROI.

Right data: Ensure the training data is accurate, trustworthy and free from bias. Make sure there are privacy, intellectual property and business rights to use the data. Generative AI models trained on flawed internet data need to be corrected by the companies using them.

Uninterrupted monitoring: AI systems must be continuously tested for seven critical things including  accuracy, fairness and cybersafety due to technical issues like model drift or model degradation. Think of it as the sensors in a car alerting the driver to potential problems. We can do this right now. This can be accomplished by embedding standards for accuracy, health and safety and other critical areas into the AI applications being built.

Supervision: Human oversight is essential. Companies must embed their values into AI systems and retain control over critical decisions by correcting their AI applications once they are alerted to deviations from their standards via the testing and monitoring described above. A swat team of trained humans is needed to correct models quickly and bring them back on track to company values.

Technical documentation: When AI models drift or fail, clear documentation will allow teams to diagnose issues and restore functionality — or deactivate the system entirely