A project team uses an AI defect-detection tool designed to flag possible cracks in concrete images. Because the tool is usually accurate, they stop doing human review and automatically reject subcontractor work whenever the AI reports a defect.

A company originally installs AI site cameras for safety alerts, but later reuses the same camera data to track individual workers’ productivity and discipline workers, even though this was not the stated or validated purpose.

A company collects workers’ safety observation reports for hazard prevention, but later uses the same reports to create individual “risk scores” for workers and decide who should receive fewer shifts. This use was not disclosed to workers.

A site team uses an AI-generated toolbox talk without checking it. The LLM includes a safety instruction that does not match the actual excavation condition, but the team still follows it because “the AI said so.”

A site supervisor uses an AI hazard-detection camera to receive alerts about workers entering exclusion zones. The supervisor still checks the site condition before deciding whether to stop work.

A contractor uses an LLM to draft a toolbox talk based on project-specific safety notes. The site manager reviews the draft, checks whether it matches the actual site condition, and edits it before sharing it with workers.