Responsible AI Practices: Security

Security

AI systems should be secure.

Security entails ensuring that AI systems behave as intended, regardless of how attackers try to interfere. When building AI solutions, security vulnerabilities tend to be neglected. In fact, security is very often a blind spot.

The security of AI systems comes with its novel set of challenges compared to traditional software systems. For instance, all the possible scenarios cannot be predicted ahead of time, especially when we are talking about problems that are difficult for humans to solve. Some examples of threats are the poisoning of training data and model theft.