While software has always been subject to security risks, with AI those risks emerge through the possibility of data manipulation. Three classes of data have an impact on prediction machines: input, training, and feedback. All three have potential security risks.
