Canada's Sovereign AI: Securing Against Manipulation

Canada is investing billions in sovereign AI infrastructure, but securing these systems presents a new challenge. While traditional cybersecurity focuses on preventing breaches, AI security needs to address how these systems can be persuaded or manipulated. Here's the thing: AI is probabilistic, meaning it can give different answers to the same question. This makes it fundamentally different from traditional software. Attackers might not need to hack an AI system; they could simply persuade it to act in unintended ways. We've already seen instances where chatbots were convinced to make absurd agreements, like selling a $70,000 car for $1. The bottom line is that as AI becomes essential to our daily lives, from national security to public services, ensuring these systems behave as intended is crucial for trust and safety.