A framer I know in Sacramento stopped wearing his phone on the job last fall. He did not lose it. He left it in the truck. He had noticed that the camera on the northeast corner of the site, the one with the blinking green LED and the little solar panel angled toward the parking lot, seemed to trigger more supervisor visits on days when his crew took long breaks. He could not prove it. Nobody told him it was happening. But the camera went up in August and by October the site super was walking over with a clipboard every time the crew sat down for more than fifteen minutes, asking if everything was on schedule, and the framer decided the coincidence was consistent enough to stop carrying a device that might be helping the system find him.
He might be paranoid, or he might not be.
A $1.2 Billion Eye in the Sky
Construction cameras are no longer dumb boxes that record grainy time-lapse footage for the owner's weekend slideshow. The construction camera market hit $1.2 billion in 2024 and is projected to reach $3.2 billion by 2033, growing at 11.4 percent annually. That growth is driven almost entirely by AI. Cameras now ship with onboard processors running computer vision models trained on millions of construction photographs, and those models do two categories of work that are technically distinct but operationally inseparable.
The first category is safety, and the case for it is strong. DroneDeploy's Safety AI, launched in October 2024 and now deployed on hundreds of U.S. sites, analyzes daily reality-capture imagery and flags conditions that violate OSHA regulations with what the company claims is 95 percent accuracy. Missing guardrails, unprotected floor openings, workers without hard hats near overhead hazards. TrueLook built an AI-powered PPE detection system on Amazon SageMaker that identifies hard hats, high-visibility vests, gloves, and protective eyewear from fixed-position site cameras. Leica's Xsight360 runs models refined over 700,000 hours of real-world construction operation to detect people and vehicles in equipment blind spots. These tools find hazards and they save lives, because in an industry where 1,034 workers died in 2023 and falls remain the leading cause of death at 38 percent of fatalities, the safety argument is not theoretical. It is 1,034 funerals a year.
Category two is productivity. And this is where the conversation gets uncomfortable, because the same camera hardware, often the same cloud platform, and sometimes the same AI model can simultaneously track how long each worker spends at each station, how many minutes pass between task completions, who is moving and who is standing still, and whether the crew took 12 minutes for lunch or 22. DroneDeploy does not just sell safety detection to general contractors. It sells "progress tracking" from the same site imagery. TrueLook generates "actionable AI insights" including "safety indexes, dashboards, and reports" for managers. A 2026 study from National Taiwan University published in Automation in Construction demonstrated an AI framework that recognizes individual worker actions, crew-level collaboration, and overall site-level productivity from ordinary construction video. Hammering, pouring, rebar placement, formwork installation. The system measures teamwork, names each task, and quantifies its duration.
Nobody asks the workers whether they want their hammering speed scored, and nobody has to.
No Firewall
Here is the structural problem. Once a camera is bolted to a pole on a residential job site, there is no technical barrier, no regulatory requirement, and no industry standard preventing the data from flowing into both a safety dashboard and a productivity dashboard simultaneously. A GC who buys a camera subscription to detect PPE violations gets the same video feed that could power crew-level productivity analysis. Whether the GC chooses to run both algorithms is a business decision, not a legal one, because no federal law restricts it, OSHA has not addressed it, and no state construction code mentions it.
Compare this to a different industry, one that already fought this battle. Warehouse workers at Amazon have been tracked by productivity algorithms for years, and the backlash has been fierce: congressional investigations, state-level legislation in California and Washington, and a growing body of research linking algorithmic management to worker injuries from pace pressure. Construction has imported the same technology without importing any of the regulatory conversation that followed.
In May 2026, the New York Times Tech Guild filed the first major union grievance over AI surveillance tools after management used a developer experience platform called DX to generate individual productivity metrics and initiated disciplinary meetings citing weekly output figures. The guild argued that the tools constituted unauthorized monitoring under their collective bargaining agreement. Construction workers, 87 percent of whom are non-union in the residential sector according to BLS data, have no equivalent mechanism to push back.
Who Gets Watched
Construction is not a racially homogeneous industry. Hispanic and Latino workers hold approximately 3.5 million construction and extraction jobs in the United States, representing roughly 30 percent of the workforce, according to Bureau of Labor Statistics data. Foreign-born Hispanic workers make up 8.2 percent of the total U.S. workforce but account for 14 percent of all workplace fatalities, a disparity driven heavily by construction, where falls, slips, and trips are the leading cause of death for this population.
These workers are disproportionately employed by small residential subcontractors with limited HR infrastructure. They are the least likely to receive written notification about surveillance systems, the least likely to have union representation, and the most likely to face language barriers that prevent them from understanding consent disclosures even when those disclosures exist. A 2024 survey by the UK's Trades Union Congress found that Black, Asian, and minority ethnic workers were nearly twice as likely to report wearable location tracking on the job (8 percent versus 4 percent for white workers) and far more likely to report that all their activities were monitored (33 percent versus 19 percent). No equivalent survey exists for U.S. construction, which is itself a data gap worth noting.
The Consent Gap
California's Privacy Protection Agency is drafting rules that would require businesses to inform job applicants and workers when AI is in use and allow them to opt out of data collection on the job without consequence. If finalized, California would become the first state to enact such protections. The California Labor Federation, led by president Lorena Gonzalez, is simultaneously pushing legislation to protect workers' privacy in break areas and require notice when employers collect performance data. "It's the old boss with new tools," Gonzalez told CalMatters in January 2025.
Meanwhile, the federal picture remains almost entirely blank. In Mobley v. Workday Inc., a federal court allowed disparate impact claims to proceed against a hiring AI vendor, establishing that automated systems face the same anti-discrimination standards as human decisions. But that case addressed hiring, not on-site monitoring. OSHA's mandate covers physical safety hazards, not digital surveillance. The Equal Employment Opportunity Commission has issued guidance on AI in hiring but nothing on AI in workplace monitoring. A Bloomberg Law analysis from May 2026 described the regulatory landscape as "a patchwork" that "leaves big employer gaps." For residential construction, where most workers are employed by subcontractors rather than the GC who installed the camera, even basic questions of who bears notice obligations remain unanswered.
NABTU and Microsoft announced an expanded AI training partnership in April 2026, providing no-cost AI literacy courses to construction trades workers through LinkedIn Learning. The program trained 1,500 instructors and launched industry-recognized credentials. Nowhere in the announced curriculum does it address workers' rights regarding AI monitoring on job sites. The program teaches trades workers to use AI. It does not teach them what AI is doing to them.
What This Means If You Are Building a Home
Ask your general contractor whether they use AI-enabled cameras on site. If the answer is yes, ask three follow-up questions. First: what data does the system collect beyond safety compliance, and who has access to it? Second: are subcontractor crews notified in writing, in their primary language, that their movements and productivity may be tracked? Third: is there a written policy separating safety data from productivity data, or does the same dashboard display both?
If your GC cannot answer those questions, they are not unusual. They are normal. Almost no residential builder has a camera data policy because the technology arrived faster than the governance framework around it. But the question matters because the labor shortage that delays your home is not purely a supply problem. It is also a demand problem. Workers choose which sites to show up to, and word travels fast about which GCs run tight surveillance. A superintendent in Denver told me last month that he lost two framers to a competitor who "doesn't have all the cameras," and he was not sure whether the cameras actually caused the departure or whether the perception was enough. In a market short 500,000 workers, perception is cost.
Limitations of This Analysis
No published survey quantifies how many residential GCs use AI camera footage specifically for productivity monitoring versus safety-only purposes. Most academic research on AI workplace surveillance focuses on warehouse, logistics, and tech-sector workers, not construction. The Pew Research data cited here was collected in 2023, before the current wave of AI-enabled construction cameras reached significant market penetration. The UK TUC data on racial disparities in surveillance may not translate directly to U.S. construction demographics due to differences in workforce composition, union density, and regulatory frameworks. DroneDeploy's claim of deployment on "hundreds" of sites is a company statement, not an independently verified figure. I did not interview workers currently under AI surveillance for this article because the legal ambiguity of their monitoring status made informed consent to participate in journalism about surveillance uncomfortably recursive.
The Strongest Case For the Cameras
1,034 workers died in 2023, and falls killed 393 of them. Every one of those deaths happened on a site where a human being with functioning eyes failed to notice or failed to act on a visible hazard. DroneDeploy's Safety AI claims 95 percent accuracy on OSHA violation detection. If an algorithm catches a missing guardrail at 2:14 p.m. that would have killed a 23-year-old apprentice at 2:47 p.m., the privacy trade-off is not close to being a close call, and anyone who argues otherwise has never attended a construction funeral. Safety cameras have a legitimate, urgent, life-saving function. The question is not whether to use them. The question is whether the same system that protects workers should also grade them, and who gets to decide, and whether the workers whose lives and labor are being captured have any say in the answer at all.
← Back to AI Home Building