Artificial intelligence (AI) has become a major part of our lives, from medical treatments to smart home assistants. But with its widespread use comes the responsibility to ensure our AI systems are secure. It’s an issue Stanford and Georgetown tackled in their new report on the security risks of AI, particularly adversarial machine learning.
The report recommends AI security be included in the cybersecurity programs of developers and users, and that organizations build AI models with security in mind. It calls for collaboration between cybersecurity practitioners, machine learning engineers, and adversarial machine learning researchers. And it suggests setting up a trusted forum for incident information sharing to ensure malicious attacks or vulnerabilities are quickly identified and shared.
Jim Dempsey, one of the workshop organizers, wrote a blog post on the report. He emphasizes the need to treat AI security as a subset of cybersecurity and to apply vulnerability management practices to AI-based features. He also notes the importance of consulting with those addressing AI bias, since AI vulnerabilities may be more analogous to algorithmic bias than to traditional software vulnerabilities.
AI security is a growing concern, but with the right steps, organizations can ensure their AI systems are secure and their users are safe. To do this, organizations should include AI security concerns in their cybersecurity programs, create a risk management framework that addresses security throughout the AI system life cycle, collaborate with AI fairness researchers, and establish information sharing among AI developers and users.
• AI security should be included in cybersecurity programs
• Organizations should create a risk management framework that addresses security throughout the AI system life cycle
• Collaboration between cybersecurity practitioners, machine learning engineers, and adversarial machine learning researchers is necessary
• Establish information sharing among AI developers and users
• Consult with those addressing AI bias to ensure AI vulnerabilities remain secure