JobRiskAI

Health and Safety Engineers, Except Mining Safety Engineers and Inspectors

SOC 17-2111Architecture & EngineeringData vintage 2026-07
Elevated exposure AI applicability score 0.240, higher than 77% of the 785 occupations measured · #8 most exposed of 35 in Architecture & Engineering

The routine layer of this job is compressing.

A substantial share of this occupation's activities overlaps with what AI already does well. Expect the routine layer to compress first, and the human-anchored layer to gain weight and value.

Where this occupation sits among all 785 measured. Bands are quintiles of the real distribution. Applicability measures task overlap with observed AI usage. It is not a probability of job loss.

What this job involves

Promote worksite or product safety by applying knowledge of industrial processes, mechanics, chemistry, psychology, and industrial health and safety laws. Includes industrial product safety engineers.

Where AI overlaps this work, activity by activity

The occupation's most important work activities (O*NET weights, normalized to this set), each with its measured AI performance across 200,000 real conversations: how often AI is used for it, how well it completes it, and how much of the activity it covers.

Work activityShare of roleAI performance
Advise others on workplace health or safety issues.16.1%Not observed 0.00
Research technology designs or applications.14.9%High overlap 0.60
Investigate organizational or operational problems.14.0%Moderate 0.57
Maintain current knowledge in area of expertise.10.7%High overlap 0.61
Develop safety standards, policies, or procedures.8.0%Not observed 0.00
Document technical designs, procedures, or activities.7.3%High overlap 0.61
Teach safety procedures or standards to others.6.9%Not observed 0.00
Inspect facilities or equipment.5.6%Not observed 0.00
Communicate with others about specifications or project details.4.6%Emerging 0.38
Present research or technical information.4.3%High overlap 0.75
Maintain operational records.4.2%Not observed 0.00
Explain technical details of products or services.3.4%High overlap 0.81

"Not observed" is a measurement, not missing data: that activity does not meaningfully appear in the usage sample. AI performance = completion × scope × coverage (user-goal view), 0–1.

What drives this score: the biggest AI-overlapped activity in this role is research technology designs or applications; the most important activity AI is not observed doing is advise others on workplace health or safety issues. The plan below is built around exactly that split.

Move first: compress the routine before it compresses you.

Next 30 days

  • Take the exposed layer yourself. Your most AI-overlapped activity, Research technology designs or applications, is exactly what a well-tooled colleague will accelerate. Be that colleague first.
  • Time-box the routine. Measure how long your repeatable tasks actually take; that's the budget AI adoption will compete against, and the time you will reinvest.

Next 90 days

  • Double down where humans hold. Grow the share of your week spent on work like Advise others on workplace health or safety issues, measured as outside AI's current reach, and take on more of it explicitly.
  • Build one AI-assisted workflow end-to-end for your team and own its quality. Being the local standard-setter compounds.

The longer game

  • Position at the boundary. The valuable seat in partially-exposed occupations is the translator: someone fluent in both the domain and the tools. Certifications matter less than a demonstrated workflow portfolio.
  • Recheck yearly. Elevated-band occupations move: new tools shift specific activities fast. The data vintage is printed at the top of this page.

Compare this occupation with another

Sources for this page:
  • AI applicability score and per-activity metrics: Tomlinson, Jaffe, Wang, Counts & Suri, Working with AI: Measuring the Applicability of Generative AI to Occupations, Microsoft Research 2025 (arXiv:2507.07935), data CC BY 4.0.
  • Occupation description and activity structure: O*NET 30.0 Database, USDOL/ETA, CC BY 4.0.
  • Bands, personalization weights and action plans are JobRiskAI editorial layers; method and limitations on the methodology page.