Facing the Two-Edged Sword of AI

Join us as thought leaders from government and industry discuss emerging best practices around the safe use of AI and the business infrastructure needed behind it.

The use of artificial intelligence (AI) is introducing a paradigm shift in cybersecurity. On the one hand, AI is a powerful asset for cyber resilience, with its ability to automatically update defenses and quickly identify and respond to multi-pronged attacks. On the other hand, bad actors have quickly adapted AI to launch hostile tactics and precision-targeted attacks, such as phishing, spam and malware or mimicking voices in messages and phone calls, and increasing agencies’ attack surfaces.

Furthermore, AI is still a very young technology; it can behave unpredictably, expose unauthorized data, and create both operational and reputational risks. Incorporating AI into agency systems requires eliminating blind spots and securing data sources, models, and infrastructure, such as preventing the AI model from pulling unauthorized data because existing security measures are insufficient. Ensuring visibility and control of AI activity are essential to harnessing its benefits while minimizing its risks.

Learning Objectives:

• Learn how to identify inadequate security guardrails for AI models and the steps needed to strengthen them
• Review common pitfalls of AI deployments, such as the inability to segment and contain anomalous behaviors, poor visibility into the AI model’s behavior, and lack of control over data flow
• Outline methods of restricting AI access to approved data sets only
• Delineate ways to trigger alerts and blocking AI access to external or unintended data sources