Artificial Intelligence (AI) Systems are susceptible to novel vulnerabilities, which can be experienced by unsuspecting users or exploited by malicious actors. AI red teaming can enhance the protection of AI systems throughout the nation.
AI Red Teaming: Advancing Safe and Secure AI Systems
Download Resources
AI’s potential as a powerful dual-use technology means its adoption will be swift and pervasive throughout industry and government. AI’s rapid and widespread adoption will be paced by the speed of AI’s development and advancement, which is unprecedented. These factors together will create an environment where AI systems in use by government and industry possess a broad “attack surface,” vulnerable to the malign actions of our nation’s adversaries at the state and non-state levels.
This paper identifies benefits that adoption of recurring AI red teaming efforts can have for AI applications, in their development, deployment, and use. This paper proposes measures that can be undertaken for advancing whole-of-nation AI assurance against global adversaries, safeguarding the public, protecting critical infrastructure, and ensuring the continuity of mission deliveries by the United States government.