Construction AI Pilot Tracks Use, but Injury Prevention Evidence Remains Thin

Construction and infrastructure operators are using artificial intelligence to flag job-site hazards, but a three-month Amey pilot tracked tool use rather than injuries, leaving its safety effect unmeasured.

FYLD announced a rollout of its AI-enabled workflows across Amey’s Highways division, with the program intended to cover more than 2,500 field workers. The platform analyzes short videos and other field information to identify safety, quality and delivery risks. Supervisors can use the results to review work remotely.

During the three-month pilot, the system logged more than 500 jobs. An AI-enabled assessment was recorded for 95% of the jobs; 85% of users were consistently active. No injury-rate results were reported for the pilot, so the figures describe adoption rather than whether the tools reduced injuries.

A systematic review published July 11, 2025, examined workplace AI and safety research published from 2018 through April 2024. Researchers screened 1,255 articles, and two met the review’s eligibility criteria. One was rated medium quality and one low quality; none was rated high quality.

The better-rated study involved 94 white-collar workers in Japan and tested a chatbot for musculoskeletal symptoms. It did not study a system that detects hazards at construction sites. The review may not capture company-held efficacy findings that are proprietary or unpublished, and its search period does not cover later deployments.

There were 25 construction worker deaths in Great Britain in 2025/26, compared with an annual average of 37 across the five years from 2021/22 through 2025/26. The 2025/26 count is provisional and will be finalized in July 2027. The figures do not show whether AI affected safety outcomes. Fatality rates are more useful for comparisons because workforce size changes over time.

Waleed Umer, an assistant professor at Northumbria University, said AI should support human oversight. “Treat AI as an extra pair of eyes, not a replacement for supervision. And be clear about what it can’t see.”