Hi, It is more & more common that some "managers" in companies are using AI to create their own risk assessments for the workplace. All of the assessments I have been asked to review go into lots of detail but the basic principals of obvious and forseeable hazards. I wonder if the HSE would be happy with company managers creating their own workplace risk assessments and passing them off as created by a competent person.

Persnally I would hope that the HSe would still expect a competent person to be involved in the creation and sign off.

What are your thoughts?

Originally Posted by: Redmund

I wonder if the HSE would be happy with company managers creating their own workplace risk assessments and passing them off as created by a competent person.

I am curious as to where you have read that risk assessments are / must be created by "a competent person".

There is a general employer duty to seek out and receive "competent advice in matters of Health & Safety".

Personally I love the idea of managers taking their responsibilities seriously and creating an RA rather than waiting for the elf n safety bod to do it for them.

As to using AI that is probably as bad as a set of "off the shelf" consultants documents, and likely no better.

How do you determine what is competence when there is actually no specified measure.

Hi Redmund

Your first thread here, so welcome to the Forums.

I am firmly of the view that the best people to lead on the process of putting together task risk assessments are those doing the work and those managing/supervising them, with a input from specialists as appropriate.

If that "competent person" is an H&S professional there is no way that they know about how a job is done and what the upsides and downsides of precautions X, Y and Z might be UNLESS they have done the job themselves.

...and even if that is the case, imagine the scenario of a crane driver turned OSH professional (nowt wrong with that) what the expertise the crane driver might have had perhaps 10 years ago is now likely to be out of date.

Long time ago, I went on a Cranes Appreciation course and got to play with an RB22 crawler crane. What seemed to be the simplest of tasks, moving a load from A to B proved to be VERY hard and each time I made a mistake I found myself overcorrecting and made the situation worse. 

On the same course I got to play with a hydraulic crane which was so much easier as, in effect, the crane did much of the thinking for me.

These days you would be very unlikely to find an RB22 crawler crane still being used in the UK, so that the incredible skill that the operator learnt those years ago has little relevance to craneage today, though the PRINCIPLES will still apply and if that operator has had a career change they should be able to add value to the risk assessment process NOT ONLY for craneage operations but other things as well.

As for AI, like its use for anything else it may have its usage in risk assessments as long as you understand the potential pitfalls. But as Roundtuit indicates, not really much different from the outputs of many an OSH Consultant (OR other OSH professional) over the years.

Edited by user 23 September 2025 14:53:28(UTC)  | Reason: Typo

Good Morning All,

I think just like with everything AI is simply a tool and personally from my experience as with Peters crane operator skill is a huge factor,

It’s a little bit like excel skill levels which seems comparable as they tend to vary hugely in workplaces and it’s been around longer. Some peoples limitations with AI is simple questions and answers, for myself it’s helped me create macros and wrote code for me with much trial and error that I would have never been able to comprehend alone. Albeit I admit I pay for a paid version which gives me more capability but I’ve also created a safety assistant, it’s been fed and stores countless PDFs from HSG65, different Acts and legislation and even books and quite a lot in between, it cross references information to the web and is instructed to reference the answer either stored via stored PDF or the web so I can check answers myself, so guess where I go when I need help?

Do I use it to write risk assessments? No. Working alone as I do, do I treat it like a colleague to have a glance at my work? Often. Would I be happy that a manager took a big enough interest in health and safety to make a risk assessment to review regardless what tool was used? Absolutely.

Wee summary supporting what everyone has said so far...two issues competency and the use of AI.  I have my own private LLM which I have used for the last 5 years only as a advisor, you still have to know when it kicks out c**p...there are also further parts to competency in CoMAH advice but lets just keep it light for now ;)

I have a summary below on both subjects which I have taken from MoJ and The Law Society guidance along with my own experience along with what I think the  

Competency: Key Principles from Case Law  

• Competence is objective: It depends on knowledge, experience, and training, not just job title.
• Specialist risks require specialist knowledge: Technical hazards (e.g., asbestos, legionella) demand assessors with relevant expertise.
• Employers are strictly liable: They cannot delegate responsibility and then claim ignorance if the chosen assessor proves incompetent.
• Competency is a continuing duty: Employers must review and update competence as circumstances change.

My personal Sudo-Legal View on AI Use in Risk Assessments

• Permissible as a tool: AI may assist human assessors by gathering data, spotting patterns, or drafting assessments.
• Not sufficient on its own: AI cannot meet the statutory requirement of “competence” under MHSWR 1999 reg. 7. A competent person must review and validate the assessment.
• Liability remains with employer: If an AI-generated assessment is flawed, the employer is still legally responsible — both civilly and criminally.
• Best practice: Employers using AI should treat it as an advisory tool, ensure human review, document the process, and confirm competence of those overseeing it.

My view on HSE's Anticipated Regulatory Approach

• HSE guidance (forward view): The HSE is likely to acknowledge AI as a tool but require that assessments are ultimately verified by competent persons.
• Standards development: The UK Government and regulators may develop formal guidance or codes of practice on AI in occupational safety (similar to current HSE-approved codes).
• Risk of enforcement: If harm occurs and AI assessments were not human-verified, the HSE will likely treat this as a breach of duty under HSWA 1974.

https://www.hse.gov.uk/news/hse-ai.htm

Practical Recommendations

• Human sign-off: Ensure all AI-generated assessments are validated by a competent person under reg. 7.
• Audit and transparency: Keep full records of AI outputs, human review, and final approval.
• Risk hierarchy: For high-risk or technical hazards (e.g., asbestos, confined spaces, hazardous substances), rely on qualified human specialists; AI may assist but cannot lead.
• Governance framework: Develop an internal policy setting out when and how AI may be used in the risk assessment process.
• Prepare for regulation: Monitor HSE publications; adopt best practice early to demonstrate compliance and good faith if challenged.

Exclusive reliance on AI would contravene the HSWA 1974 and MHSWR 1999. AI can be deployed as a valuable aid but must never replace competent human judgment. I would be advising companies to adopt a cautious approach, with human sign-off as a legal and practical safeguard.

To summarise:

Your client is using AI to write RA's then expects you to sign the document as having created it.

The HSE having seen your signature which given the poor output from AI has questioned your competence and qualification.

Personally I would be re-writing their RA process defining who was doing what and if you have to sign documents to be paid documenting on what basis you have signed - author, sense check etc.

It is long overdue giving AI the credit for its output.

Originally Posted by: Redmund

The reason I said that Risk assesments should be carried out by a competent person is that recently a client was visited by the HSE and asked if I was qualified and competent to write and sign off the risk assessments.

My experience so far with AI is that it misses out on the skills, knowledge and experience of people including H&S advisors and some people might think that AI can replace an experienced and qualified H&S advisor, which would leave them to issues with the HSE.

mmm...wee narrative change here...so all the HSE did was ask a question of competency in regards to R7.  The trap was the risk assessment - you facilitate and the client representative always signs off as it is their legal duty...as they would have expected your to answer that way and that you fullfill the requirements of R7 for skills, knowledge and experience...AI is a red herring here...and it can never replace a competent advisor, but can provide assistance.  

The trouble with 'AI' (having trained my own) is that most people use it like google and accept the first answer rather than seeing it as the tool it can become...it is still a child and needs educating, it also has some deveoping social behaviours i.e. it makes s**t up to answer a question (I call it the FaceBook Phenomena :)..as it uses data from there so thinks that behaviour is ok)...barristers were recently repremanded by a judge for using AI when it created case law that didn't exist...

I do worry about when it becomes sentient and realised I'm the ar**ole... :)

The issue I see with AI is its only as good as the query inputted, A RA written for task A on a site in a rural location will often be different to the essentially  the same task in a urban location  and could well even evolve in the same location as a project progresses.  If the person inputting the information has all the knowledge about the process, controls and legislation then he’s probably skilled enough to write it himself. 

I do use ChatGPT to for instance write job descriptions for indeed but I’ve never managed to generate one that has not needed some amending.