The U.S. National Security Agency teamed up with other stateside departments as well as some in the U.K. to issue a cybersecurity information sheet, or CSI, on artificial intelligence system development.
The guidance, which includes input from the U.K. National Cyber Security Centre and U.S Cybersecurity and Infrastructure Security Agency, warns against adversarial machine learning attacks that could compromise data in AI technologies.
The CSI emphasizes the implementation of secure design, development, deployment and operation in AI systems. It recommends conducting a holistic assessment of AI-specific threats and using the findings to prepare the system for the evolving attack vectors.
The document also urges developers to compare the benefits and weaknesses of different AI models, taking into consideration factors such as training dataset characteristics, use case appropriateness, model complexity and component supply chains.
“We wish we could rewind time and bake security into the start of the internet. We have that opportunity today with AI. We need to seize the chance,” NSA Cybersecurity Director and Wash100 awardee Rob Joyce commented.