Step 4
Implementation of an AI governance framework
A robust AI governance framework is essential for ensuring that AI systems are deployed responsibly and comply with regulatory standards.
Tailoring the framework to the level of AI risk and organisational role allows for effective oversight and minimises potential liabilities. Below is an overview of obligations that may apply based on the risk classification of AI systems, building on the assessments in steps 1 and 2.
1. High-risk AI systems: comprehensive governance and compliance obligations
For high-risk AI applications (those that could significantly impact individuals, society, or business operations) a rigorous set of governance measures is required to mitigate risk and maintain accountability. Key obligations for deployers of high-risk AI systems may include:
Transparency obligations
Reporting and cooperation with authorities
Automatic recording of events
Quality management system
Technical documentation
Human oversight
Conformity assessment
Post-market monitoring
Risk management system
Data requirements
Registration
Cybersecurity, accuracy, and robustness
Transparency obligations
Reporting and cooperation with authorities
Automatic recording of events
Quality management system
Technical documentation
Human oversight
Conformity assessment
Post-market monitoring
Risk management system
Data requirements
Registration
Cybersecurity, accuracy, and robustness
Deployers are subject to less intensive obligations
For deployers less intensive obligations apply including AI literacy, human oversight, data governance and transparency.
AI literacy
Train relevant staff to understand the basic principles, regulatory landscape, and ethical implications of AI, fostering a culture of responsible use.
Human oversight
Ensure a degree of human oversight to monitor AI outputs and provide intervention capabilities if needed, maintaining accountability.
Data governance
Establish clear data governance practices to ensure the accuracy, security, and fairness of data inputs, reducing potential bias or errors.
Transparency
Provide a basic level of transparency regarding AI functions and objectives, especially when engaging with end-users or customers.
AI literacy
Train relevant staff to understand the basic principles, regulatory landscape, and ethical implications of AI, fostering a culture of responsible use.
Human oversight
Ensure a degree of human oversight to monitor AI outputs and provide intervention capabilities if needed, maintaining accountability.
Data governance
Establish clear data governance practices to ensure the accuracy, security, and fairness of data inputs, reducing potential bias or errors.
Transparency
Provide a basic level of transparency regarding AI functions and objectives, especially when engaging with end-users or customers.