Certifying AI Systems

The Department of the Air Force is employing urgent strategies to enhance its capabilities for Great Power Competition by investing in AI; however, all AI models are subject to financial, regulatory and security risks without proper monitoring and controls in place.Ā There is a looming threat of bad actor models that output inaccurate predictions and reduce operational effectiveness as well as introduce vulnerabilities to adversarial exploitation and result in the complete loss of trust and credibility in the AI investments. The first step in addressing these vulnerabilities is to have full visibility and the ability to control these risks through AI specific governance, enabling NIST policies and principles such as Visible, Accessible, Understandable, Linked, Trustworthy, Interoperable and Secure (VAULTIS) in order to Certify AI Systems. However, there is a hesitation to comprehensively monitor and govern AI because there is a fear that these processes are inadvertently impeding innovation given that most governance is manual, inconsistent and not scalable.