The human side
of artificial intelligence.
AI will not only help us do our work. It will increasingly judge how responsibly we do it.
A different safety question.
AI safety and security are essential concerns. AI for HSE examines a complementary challenge: how well can AI analyze, evaluate, and support the safety of people and the environment?
As agentic AI enters business operations, models may influence whether a task is considered safe, a hazard is escalated, or an environmental obligation is recognized. An articulate answer is not enough. These judgments need evidence.
What we are building
AI for HSE is the research initiative behind AI4HSE-Bench, a benchmark family focused on health, safety, and the environment. The current release, AI4HSE-Bench v1, tests multi-select question answering under a fixed protocol, with separately reported answer-only, reasoning-assisted, and structured-output tracks.
It is a starting point for measuring domain knowledge and answer reliability. It does not yet test autonomous agents in operational environments or establish readiness to make real-world safety decisions.
Our publishing principles
- Report failures alongside correct answers.
- Make methods and limitations visible.
- Keep incompatible evaluation tracks separate.
- Publish aggregate evidence while protecting the closed question set.
The current dataset contains 100 questions. Public aggregates show model scores, response coverage, evaluation settings, and limitations. Models are added one at a time as their evaluations are completed.
Read the AI4HSE-Bench v1 methodology →See what we measure today and the questions proposed for future studies in the Research Programme.
Research with industry
We welcome conversations with organisations interested in supporting independent research through funding, expert review, and practical HSE questions. Support cannot influence rankings or publication decisions.
Industry Research Partners →Led by Pavel Kosyrev
AI for HSE is an independent research project led by Pavel Kosyrev. His current work focuses on AI for healthcare and for health, safety, and the environment (HSE).
Educated in the United Kingdom, Pavel holds a degree in architecture. His work on interactive environments developed his skills in programming and prototyping. He subsequently joined his family’s occupational health and safety business and gained qualifications in occupational, industrial, and fire safety, environmental protection, first aid, and occupational health.
He founded NOVA in 2016 to develop technology for health and safety, working across hardware and software. AI for HSE brings that engineering and industry background to the evaluation of AI knowledge and answer reliability.
For research questions, expert review, or a conversation about industry collaboration, contact Pavel directly.