Self-Assessment
Is Your Plant Ready for AI?
15 Critical Questions Every Operator Should Answer
If you can't answer "yes" to these questions, your facility has AI governance gaps that need immediate attention. "I don't know" counts as "no."
Know Your AI
Do you have a complete inventory of every AI/ML system running in your facility — including vendor-embedded analytics?
You can't manage what you don't know about. Vendor packages often include AI you never asked for.
Ref: OSHA PSM — Process Safety Information; NIST AI RMF GV-1
For each AI system, do you know whether it can directly change process parameters — or only make recommendations?
An AI that adjusts a setpoint is fundamentally different from one that suggests an operator adjust it.
Ref: ISA-84; OSHA PSM — Operating Procedures
Manage the Change
Are AI deployments and model updates going through your Management of Change process?
AI changes process behavior. That's a change. Period. A retrained model can behave completely differently — same interface, different decisions.
Ref: OSHA PSM — MOC; CCPS MOC Guidelines
Can your vendor push a model update without your knowledge or approval?
If the vendor can force an update you haven't validated, you've lost control of your process.
Ref: OSHA PSM — MOC; NIST AI RMF GV-6
Protect Your People
Do your SOPs tell operators what to do when the AI recommends something they disagree with?
Operators need clear authority to override. "Use your judgment" isn't a procedure.
Ref: OSHA PSM — Operating Procedures
Can every operator on every shift override AI systems quickly and without hesitation?
If an operator hesitates because they're not sure how to override, you've failed at training.
Ref: OSHA PSM — Training, Operating Procedures
Do operators understand what each AI system on their unit does — and what it doesn't do?
An operator who thinks the AI is watching something it's not is a disaster waiting to happen.
Ref: OSHA PSM — Training
Guard Your Safety Systems
Is AI architecturally separated from your Safety Instrumented Systems?
Defense in depth means AI can't touch SIS even if compromised. Full stop.
Ref: IEC 61511; ISA-84; IEC 62443-3-3
Has your SIL verification accounted for AI system interactions?
If AI can affect demand rate on a Safety Instrumented Function, your SIL calculations may be wrong.
Ref: IEC 61511; ISA-84
Prepare for the Worst
During an emergency shutdown, do AI systems fail to a safe state automatically?
If AI tries to "help" during an ESD, it can interfere with emergency procedures.
Ref: ISA-84; OSHA PSM — Operating Procedures
Can you reconstruct what AI systems recommended or did in the hours before an incident?
If you can't replay the AI's actions, your incident investigation has a hole in it.
Ref: OSHA PSM — Incident Investigation; NIST AI RMF MS-4
Know Your Risks
Has your PHA specifically considered AI system failure modes as initiating events?
Your PHA probably didn't consider "AI gives confidently wrong recommendation" as a deviation. It should.
Ref: OSHA PSM — PHA; CCPS LOPA Guidelines
Are AI systems included in your IEC 62443 zone and conduit model?
If AI systems aren't in your security architecture, they're unprotected attack surfaces.
Ref: IEC 62443-3-2
Stay Accountable
Could you show an OSHA inspector how you govern AI systems in your plant — today?
If you can't demonstrate it, you don't have it. Inspectors don't accept "we're working on it."
Ref: 29 CFR 1910.119; OSHA PSM NEP
After a turnaround, do you verify all AI systems are back to their pre-turnaround state — or properly updated?
Turnarounds reset things. AI configurations get lost, data connections break, and nobody checks until something goes wrong.
Ref: OSHA PSM — PSSR
Score Yourself
10+ "Yes" answers: You're ahead of most facilities. The full 70-question assessment can help you close remaining gaps.
Fewer than 10: You have significant AI governance gaps. Start with questions 1, 3, and 8 — inventory, MOC, and SIS separation. Everything else builds from there.
© 2026 Council for Industrial AI Safety. Free to distribute with attribution.