Many AI applications combine cloud computing with edge devices. Should data processing take place on the edge or in the cloud? The Arm 2020 Global AI Survey suggests that consumers prefer that data processing happens process locally and that data be uploaded to the cloud only when absolutely necessary:
Once data is in the cloud, what security challenges concern those who are responsible for all that consumer data? According to the 2020 Cloud Security Report (a survey conducted by Cybersecurity Insiders), Misconfiguration of the cloud platform (68%), Unauthorized access (58%) and Insecure interfaces/ APIs (52%) were perceived to be the top three security threats when running applications on a public cloud.
Security, machine learning, and systems researchers are collaborating to develop tools that can help companies build secure and privacy-preserving AI and machine learning applications. Research centers like RISELab, and companies like Apple and Google, are producing new tools that organizations are starting to deploy. With that said, we are still in the early days of deploying ML systems that combine cloud platforms and edge devices, so I expect even more security breaches and new attacks against ML in the near future.