From tinkering to trust: how safety leaders can adopt AI responsibly

Across many organisations, AI is increasingly becoming part of how health and safety work gets done.

A 2026 survey of EHS leaders by Wolters Kluwer and the National Safety Council found that 82% already use AI in some form, and only 2% use none at all. The same survey found that 90% had at least one concern about it. Almost everyone is using AI in health and safety, and almost everyone is uneasy about how. That gap between adoption and confidence is the real challenge for safety leaders.

This curiosity to use AI to streamline processes is not a bad thing. In fact, it shows that safety professionals are already looking for ways to reduce administrative burden, communicate more effectively and spend more time on the work that really requires their judgement and expertise.

As AI becomes more accessible, safety leaders need to think less about whether it has a role and more about the conditions needed for safe, responsible and effective use. 

For many safety teams, AI adoption starts with experimentation. Photograph: iStock

From experimentation to adoption

For many safety teams, AI adoption starts with experimentation. Someone tests whether a tool can summarise a document, rewrite it in plain English, or turn rough notes into a more usable first draft. These small use cases can be genuinely useful, but they are only the starting point.

The greater opportunity lies in moving from individual experimentation to responsible adoption. That does not mean safety leaders need to become AI experts, but they do need to become AI-literate leaders who can govern and apply AI responsibly. 

In practical terms, AI literacy means being able to ask the right questions, such as:

• What is the tool being used for?
• What information is being entered into it?
• How reliable is the output?
• What context might be missing?
• Who is checking the output before it is used?
• What could go wrong if the output is incomplete, inaccurate, or applied in the wrong setting?

As AI becomes more embedded in work, safety leaders have an important role to play in making sure it supports better decision-making, rather than creating new risks through unmanaged use.

The most effective leaders will be those who combine curiosity with control – exploring the potential of new tools with a critical eye, while avoiding both uncritical adoption and blanket rejection based on caution or principle.

Building confidence through clear AI governance

It is tempting to think of AI governance as something complex, legalistic or reserved for large organisations with specialist technology teams. In reality, every organisation using AI in health and safety needs some basic rules in place. 

Good governance creates the conditions for AI to be used with confidence, rather than uncertainty.

Without clear boundaries, people may enter sensitive incident details into unapproved tools, rely on outputs that have not been checked, or use AI for tasks where human judgement is essential. Different teams may use different systems in different ways, making it difficult to understand where risk lies. Over time, informal use can become normalised before the organisation has properly considered the implications.

As AI develops, governance also needs to account for systems that do more than respond to a prompt. Where AI is able to follow a process, trigger a workflow, gather information or suggest next steps, organisations need to be clear about the limits of that autonomy. The question is not only what people are allowed to do with AI, but what AI is allowed to do within the organisation’s safety processes.

A practical AI governance plan should answer some simple questions:

• Which AI tools are approved for use?
• What types of tasks are suitable for AI support?
• What data must never be entered into open 
  or unapproved systems?
• Who reviews AI-generated content before 
  it is shared or acted upon?
• Where AI is able to act within a workflow, what 
  is it allowed to do without human approval?
• Who can pause, stop, or override an 
  AI-enabled process?
• Is there a clear record of what the system did, what information it used and what it suggested?
• How are errors, poor outputs or unexpected results escalated?
  How are suppliers assessed?
• How is AI use monitored and improved over time?

Purpose-built safety AI can help reshape how information is connected, analysed and acted upon. Photograph: iStock

For safety teams, data governance is particularly important. Incident reports, investigation notes, witness statements, occupational health information and employee records can all contain personal, sensitive or confidential information. 

Organisations need to be clear about how that data is handled, where it goes, who can access it and whether it is being used appropriately.

This may sound like the less exciting side of AI, but it is also where trust is built. Clear guardrails give safety teams permission to experiment within safe boundaries.

They help protect workers, support compliance and give leaders confidence that AI is being used in a way that aligns with the organisation’s values and responsibilities. As AI becomes more capable, those guardrails also help ensure that automation remains contained, auditable and accountable.

Where AI can help today

Much of the conversation around AI focuses on what it might do in the future, but it is just as important to understand where AI can already add real value today. 

For safety teams, that value is emerging in two main ways: through general AI tools that can support everyday productivity, and through purpose-built safety AI that can work with the information organisations already hold.

The organisations getting the most from AI are unlikely to be those simply adding it to yesterday’s processes. They will be those willing to rethink where the work sits and what technology can help teams stop doing manually.

General AI can help safety teams complete familiar tasks more quickly. Many safety tasks involve turning complex information into clear, usable outputs, and AI can help speed up that process by providing a structured first draft while still leaving the final judgement with the safety professional.

This can be particularly helpful when teams need to draft or update safety policies and procedures. AI can help structure content, improve clarity, simplify technical wording and adapt existing material into a more accessible format. 

Of course, these outputs still need to be checked carefully. Accuracy, context and operational relevance are all important, particularly where information could affect how work is planned or carried out. But general AI can provide a useful starting point, especially for busy teams who need to turn detailed information into clear, practical guidance.

Purpose-built safety AI can go further by working within the context of the safety management system itself. It can help reshape how safety information is connected, analysed and acted upon. This is important because safety teams are not just creating content; they are managing live information about incidents, investigations, actions, risks and performance.

After an incident, for example, the immediate priority is to respond, understand what happened and keep people safe. At the same time, safety managers may need to brief senior leaders, prepare updates for teams, capture learning for review meetings, identify follow-up actions and ensure relevant lessons are shared.

Doing this manually can involve repeating the same information in slightly different formats, at exactly the moment when time, clarity and consistency matter most.

AI-powered incident analysis can help by using the data already captured in the incident record to create useful first drafts and prompts for review. It can summarise what happened, describe the impact, outline actions taken, highlight possible root causes, identify gaps in the narrative and suggest areas that may need further investigation.

The same underlying information can also be adapted into different outputs: a plain-language team briefing, a structured investigation summary, a tailored toolbox talk or a concise executive update. 

When incident information is structured and shared more effectively, learning can reach the right people faster. Teams can see what has happened, what may have contributed to it, what actions are being considered and where further attention may be needed. 

Purpose-built AI can also support better action management. Actions created after incidents, audits, inspections or observations are only useful if they are clear, specific and capable of being followed through.

AI-powered safety software can help identify unclear wording, prompt users to provide more detail and suggest potential corrective actions based on the information available. This gives safety professionals a stronger starting point for review and helps reduce the confusion that can come from vague or inconsistent action descriptions. 

The same principle applies to safety reporting. Many organisations hold large volumes of information across incident reports, investigation findings, corrective actions, audits, inspections, observations, risk assessments and trend reports.

Valuable signals can remain hidden, especially when safety teams are stretched. Safety-specific AI can help bring those signals to the surface by identifying patterns, highlighting possible gaps and showing where similar issues may have occurred before. 

The real opportunity is to bring organisational learning into the flow of safety work. When AI is connected to relevant safety data, it can help professionals draw insight from individual records, identify recurring themes and understand where risk, performance or priorities may be shifting. 

It is always important to remember that AI-generated insights should support, not replace, professional review. Safety teams still need to understand where the data has come from, whether it is complete, what assumptions may be present and what requires further investigation.

AI can help bring relevant information to the surface, but people remain responsible for interpreting that information in context and applying judgement.

Looking ahead: AI agents and controlled automation

While there is already plenty that AI can help with today, the next phase of development is likely to involve more active forms of support. This is where the conversation begins to move from AI as an assistant to AI as an agent.

An AI agent can be understood as a system that does more than respond to a single prompt. It can follow a process, gather information, ask questions, complete defined tasks and work within set parameters. Within health and safety, this creates some interesting possibilities, particularly where processes are structured but time-consuming. 

However, the principle has to come first: an AI agent should never determine root cause or close out an investigation on its own. A human investigator owns the judgement; the agent prepares the groundwork.

One emerging example is early-stage incident investigation support. In the future, a controlled AI agent could help with the first stage of an investigation by conducting a structured interview using an approved question set.

It could ask relevant follow-up questions based on the answers given, capture responses, summarise key evidence, highlight gaps and suggest next steps for the human investigator to review.

It could also help gather supporting information. For example, it might identify similar previous incidents, related risk assessments, open or overdue actions, recent inspection findings, relevant procedures, training records or wider trends across sites and departments.

This could be valuable because incident investigation often depends on the ability to collect information quickly, ask consistent questions and connect individual events to wider organisational learning. AI agents could help prepare that information more efficiently, allowing human investigators to focus on judgement and context.

But this is also an area where organisations would need to proceed carefully. Workers should not feel they are being monitored, interrogated or judged by a black-box system. If AI is used in investigation processes, transparency, fairness and trust will be essential from the start.

This is also heading towards formal obligation rather than good practice alone. Under the EU AI Act, employers using high-risk AI systems in the workplace will need to inform workers’ representatives and affected workers before those systems are used.

The exact timing and interpretation of some requirements is still developing, but the direction is clear: organisations using AI within higher-risk workplace processes, such as investigations, should plan for transparency from the outset.

Organisations would need to consider consent, data protection, audit trails, escalation routes, bias, accessibility and the quality of the questions being asked. They would also need to be clear about when a human investigator takes over, how AI-generated summaries or suggestions are checked, and who can pause, stop or override the process if needed.

The potential is significant, but so is the responsibility. Agentic AI could help safety teams gather and structure information more effectively, but it must be introduced in a way that supports learning and fairness, not automation for automation’s sake.

The opportunity for safety leaders

For organisations in the early stages of AI adoption, the best approach is not to wait for perfect certainty or let experimentation continue without direction. A more practical route is to start small, learn quickly, and build responsible foundations.

AI has the potential to improve health and safety work, but only if it is adopted with care. The organisations that will benefit most will be those that understand where AI adds value, where human judgement is essential, and how to build trust into the way technology is used. 

For safety leaders, the real opportunity is to become confident, curious and responsible leaders of AI-enabled safety.

For more information see:
notifytechnology.com
[email protected]
T. +44 (0)330 390 0530

Greta Salvesen is content lead at Notify Technology