Turner Construction did something unusual on its 22nd annual Safety Week: it gave the rest of the industry one of its better internal tools.
SafeT Coach, the GC’s AI-powered jobsite safety assistant, is now available free to anyone outside Turner. The tool is grounded in Turner’s Environmental Health & Safety standards, runs inside OpenAI’s ChatGPT environment, and answers plain-language safety questions on a phone. It also evaluates uploaded jobsite photos, flags hazards, and recommends controls. A Gemini version is in development.
For an industry where the safety dataset is fragmented across thousands of self-perform shops and small subs, putting Turner’s playbook inside a chat interface is a bigger deal than it sounds.
What It Actually Does on a Jobsite
SafeT Coach handles two things that traditionally require a safety officer on speed dial. A worker can type “I’m tying off to a steel beam on a 14-foot platform — is that an anchor point?” and get a response calibrated to Turner’s fall-protection standard and OSHA Subpart M. Or they can snap a photo of a scaffold, upload it, and get back a prioritized list of what’s wrong: missing toe boards, gaps in cross-bracing, no engineered drawing on file. The system ties each recommendation to a control, not just an observation.
Turner says SafeT Coach logged more than 25,000 field interactions during pilot deployment, validated through 80+ stakeholders and an outside risk-partner review before public release. The internal Turner version retains its data; the external version doesn’t transmit user data back to Turner, addressing the most obvious confidentiality concern for a competing GC or specialty contractor trying it.
Why a Top-10 GC Gives This Away
Turner isn’t running this as charity. Three things are converging.
First, safety statistics in U.S. construction have plateaued. The Bureau of Labor Statistics has reported roughly 1,000 jobsite fatalities per year for almost a decade. Standard interventions — toolbox talks, training matrices, observation programs — have hit diminishing returns. AI in the field is one of the few unexplored levers.
Second, owners are starting to write AI-safety expectations into prequal questionnaires. By open-sourcing its tool, Turner sets a de facto industry standard that maps to its own systems. Subs that already use SafeT Coach are easier for Turner project teams to manage.
Third, Turner gets free real-world feedback. Outside users won’t share data, but aggregate usage patterns and refusal cases will surface gaps in the standard that Turner can fix internally. That’s a useful feedback loop for a self-perform firm with $20B in annual revenue.
The Caveats
SafeT Coach is a ChatGPT custom GPT, which means it inherits both the strengths and weaknesses of the underlying model. The responses are confident-sounding, even when wrong. It does not replace a qualified person designation under OSHA 1926 — a fact Turner is careful about in its rollout messaging. And the photo-analysis pipeline still misses some failure modes that humans catch (occluded fall arrest connectors, shoring deflection at the millimeter scale, partial cribbing under load).
For most subcontractors, the value isn’t replacing safety personnel. It’s giving a foreman or an apprentice a fast answer when the safety officer is on another site, and not having that conversation end in a guess.
Compare this to Skanska’s and Balfour Beatty’s parallel AI safety efforts — Skanska is using AI on video feeds to detect roadwork zone intrusions, and Balfour Beatty’s tools focus on near-miss pattern analysis. Turner’s bet is the inverse: put the model in front of every worker, not just the analytics team. Background reading from our earlier coverage: construction robotics crossing from pilot to production.