EHS OS™
The Operating System for Enterprise Safety
Free guide · 23 pages · For enterprise EHS leaders
BEFORE YOU BUY THE AI MAGIC.
Every EHS platform now says it is powered by AI. Almost none of them will tell you what is actually under the hood, who is accountable when it is wrong, or what happens when the model leaves a step out of a procedure. This guide gives you the questions that get those answers and it shows why your operating model, as opposed to the vendor's demo, decides whether any of it works.
The vendor question bank
Grouped for the room: provenance, training, validation, oversight, liability, operating model fit.
A printable demo-room scorecard
One sheet per vendor. Check a box only when the answer was specific. Score out of 22.
The AI-readiness test for your own system
Four questions you answer before the vendor answers anything. One honest "no" changes the plan.
YOU'VE HEARD THE PITCH
Six phrases that should
stop the demo.
WHAT'S INSIDE
A field guide, not a think piece.
Twenty-three pages written for the person who is accountable for safety
performance across real sites with real people. Read it once and you will know
what to ask, what a good answer sounds like, and what to do when the answers
don't come.
PART FOUR
What is actually under the hood
The difference between a model trained on validated safety data and a general-purpose LLM wired in through an API — and why a missing step is far more dangerous than a fabricated one.
PART FIVE
Who owns it when someone gets hurt
Duty of care sits with the employer. Vendor liability is usually capped at fees paid. What to negotiate, log, and name before deployment rather than after an incident.
PARTS TWO & SEVEN
Where AI fits in your operating model
AI is not a layer you bolt on top. It lives inside technology alignment and integration, and its success is decided by the layers around it. Includes the four-question readiness test.
A LOOK INSIDE
Built to be used in demo's, not filed away.
Where AI actually lives in the five-layer framework
Why fluent-and-wrong is the dangerous quadrant
The tear-out scorecard, one per vendor
The four-question readiness test
FROM THE GUIDE
Two questions most vendors are not
ready for.
These two appear in full in Part Four, with what a defensible answer sounds like
and what an evasion sounds like.
VENDOR QUESTION № 1
"Are you training your AI models on high-quality safety data, or are you using general-purpose, off-the-shelf LLMs like ChatGPT or Claude plugged into your system?"
VENDOR QUESTION № 2
"Are you using bad decision data to do negative reinforcement (RLHF) training on your models, and how do you prevent poor real-world decisions from being learned as acceptable ones?"
It's easy to verify something that's wrong. It's very difficult to notice something that simply isn't there.
BEFORE YOU DOWNLOAD
Who is this written for.
Worth your time if you are
Probably not for you if you want
WHO WROTE IT
JJ Prinsloo
Creator of the EHS OS™ framework
Twenty years inside enterprise EHS and ESG implementations across mining, energy, utilities, manufacturing and consumer products — on all three sides of the table: as an implementer, inside a software vendor, and now designing the operating models that platforms have to run inside of.
This guide comes out of watching the same failure repeat for two decades, in newer and newer clothes: technology purchased before the system underneath it was ever designed. AI has raised the price of that mistake considerably.
Ask better questions than the demo prepared for.
Twenty-three pages, a printable vendor scorecard, and a readiness test you can run before your next platform conversation. Free.
1,848 weekly readers
