Using security tools and adhering to best practices can mitigate AI cybersecurity risks. There are many types of risks related to AI. Many of them are in the area of privacy or ethics (see other sections). Non-security issues include algorithmic bias, transparency, proportionality, legality, user rights and precision.
If you're not responsible for privacy, then these aspects are more for your privacy colleagues, but keep in mind that it's important to understand them, as AI privacy is a concerted effort. Papernot stated that efforts need to be made to specify machine learning security and privacy policies. He argued that researchers need to find the right abstraction or language to formalize the security and privacy requirements of machine learning with precise and unambiguous semantics. As a useful model, he mentioned a 1975 article by Saltzer and Schroeder describing 10 principles related to information protection in computer systems (see box 5.1 for the full list).
For that reason, the Papernot team developed a different approach called PATE Private Aggregation of Teacher Ensembles. 5 With PATE, the user can protect sensitive data in a dataset by dividing it into partitions, where the only requirement is that any training point be included in a single partition. A machine learning model (called a “teacher”) is trained on each subset of data, so that the result is numerous independently trained models to solve the same task using different subsets of data. Each model gets a “vote” on the correct label for a given entry.
Cybersecurity experts are increasingly concerned about AI attacks, both now and in the near future. The Rise of Offensive AI, a report by Forrester Consulting, revealed that 88% of decision makers in the security industry believe that offensive AI is on the way. Half of the respondents expect an increase in attacks. In addition, two-thirds of respondents expect AI to lead new attacks. By using AI security to protect and defend, your AI systems become more intelligent and effective with every attack. Papernot analyzed some of the ways in which adversaries can attempt to exploit artificial intelligence systems, the possible mechanisms for detecting or thwarting attacks, and how to translate existing principles for the design of secure computer systems to the design of secure artificial intelligence systems.
We often hear about the positive aspects of artificial intelligence (AI) security, since it allows us to predict what customers need through data and offer a personalized result.
