Smart buildings promise healthier air: sensors detect carbon dioxide, particulates and volatile compounds, and AI adjusts ventilation before occupants notice a problem. The same data, however, reveal a great deal about the people inside. Carbon dioxide levels show when a room is occupied and by how many; patterns over time show routines, absences and habits. An air quality system can quietly become a surveillance system.
Our paper in the Journal of Building Engineering reviewed a decade of studies on AI driven indoor air quality management and found that most optimise for prediction and control while treating privacy as an afterthought, if at all. In offices, homes and especially healthcare settings, where both air quality and confidentiality carry real risk, that is not acceptable.
We proposed a privacy preserving platform built on edge computing and federated learning. Data are processed close to where they are collected, and models learn across buildings without raw data leaving each site. At its centre is a SITA model, which tunes what is shared across four dimensions, spatial, identity, temporal and activity, before any modelling takes place. A facility manager can see that a floor needs more ventilation without seeing who sat where, and when.
The work sets out an architecture and a workflow that integrates with existing building management systems, so that privacy can be designed in rather than bolted on. It also makes a broader point about smart buildings: trust is a performance requirement. Occupants who believe they are being watched will disable sensors, and a building that cannot sense cannot respond.
As AI spreads through the built environment, the question is not whether buildings collect data but on whose terms. Privacy by design is how the answer stays with the occupants.
The work behind it
Quang, T. V., Doan, D. T., Ngarambe, J., GhaffarianHoseini, A., Ghaffarianhoseini, AH., and Zhang, T. (2025). AI management platform for privacy-preserving indoor air quality control: Review and future directions. Journal of Building Engineering, 100, 111712. Open access.