Manufacturers face Article 4 of the EU AI Act from two sides: as deployers of AI agents in planning, procurement, quality and maintenance, and sometimes as providers when AI goes into the products they sell. Either way, they must ensure staff and others acting for them have sufficient AI literacy, on the plant floor as well as in the office.
What does Article 4 ask of a manufacturer?
Article 4 applies to every provider and deployer of AI systems. In substance, they must take measures to ensure, to their best extent, a sufficient level of AI literacy among staff and other persons dealing with the operation and use of AI systems on their behalf, considering their knowledge, experience, education and training, the context of use, and the people the systems are used on.
AI literacy means the skills, knowledge and understanding to use AI systems in an informed way and to understand their opportunities, risks and possible harms. Whether that requires formal training is covered in does the EU AI Act require AI literacy training.
Manufacturers have a particular exposure. AI that is a safety component of a product covered by EU product safety law, such as certain machinery, can be high-risk, and a manufacturer placing such a product on the market may carry provider duties. Engineers designing those products need literacy of a different kind from planners using a scheduling agent.
Who does it cover?
- Production planners and schedulers working with planning agents.
- Procurement and accounts payable staff working with agents that raise purchase orders and match invoices.
- Quality inspectors reviewing agent-flagged defects.
- Maintenance engineers acting on agent-predicted failures.
- Shift supervisors overseeing agents that allocate work to operators, a high-risk use where allocation depends on individual performance.
- Product engineers embedding AI in products the company sells.
- Contractors and agency workers using the company's systems on its behalf.
- Directors accountable for safety and quality.
What does literacy mean on and off the plant floor?
Planning. A governed planning agent might read orders, stock and machine availability at depth 1, propose schedules at depth 2, and release routine work orders within bounds at depth 3. Planners need to read the provenance behind a schedule: the company's own ERP stock figure is sealed, a supplier's delivery confirmation is signed, a sales estimate is declared, and a forecast from an outside source is external. A schedule built on an optimistic declared estimate can idle a line.
Procurement and payments. Agents raising purchase orders or matching invoices meet the classic fraud: an emailed request to change a supplier's bank details. An external claim cannot authorise a payment, and payee changes should be held for a named person. Accounts payable staff must know why, and must not satisfy the hold by re-keying the instruction.
Quality. An agent flagging defects from images or sensor data will produce false alarms and misses. Inspectors need to understand both, sample the agent's passes as well as its rejects, and own the release decision.
Maintenance. An agent predicting a bearing failure is giving a probability, not a fact. Engineers decide whether to stop a line, and need to know how the agent reaches its estimate and how wrong it has been before.
Workforce allocation. Supervisors need to understand what the allocation agent optimises and override it when it is unsafe or unfair, especially for operators with restrictions.
Lapses. An agent's certificate lapses after 90 days, and a run that clears no depth drops it to depth 1 the same day. Planning and procurement then go manual, and night shifts must be ready.
How does a manufacturer meet Article 4?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Sufficient literacy for staff | Core module plus role modules for planning, procurement, quality, maintenance and supervision | Records by role and site |
| Persons acting on its behalf | Contractor and agency induction before system access | Induction records linked to access |
| Product engineers as providers | Separate training on provider duties for AI in products | Engineering training records |
| Awareness of risks and harms | Modules on supplier fraud, false alarms and stale forecasts | Assessment results |
| Kept current | Briefings at agent recertification or source changes | Briefing records linked to agent changes |
| Board literacy | Director session on where agents act and product exposure | Minutes |
A step-by-step build is set out in how to set up an AI literacy programme under the EU AI Act, and the board's side in the EU AI Act made AI literacy a board question.
What will a regulator ask?
Market surveillance authorities, product safety authorities and labour inspectorates may ask: which of your products contain AI, and who in engineering understands the duties? How are planners and inspectors trained on the agents they use? How are payee changes handled? How do supervisors review allocation decisions? What happens on a night shift when an agent drops to depth 1?
What does Article 4 not cover?
It does not set product safety, machinery or workplace safety rules, which apply in full. It does not replace provider duties for high-risk products or deployer duties for high-risk workforce agents. It does not prescribe course content or credentials. Comparisons are in Article 4 in logistics and in wealth management.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
Pryme Intelligence Academy's Builder track is for whoever designs, trains and certifies agents, about 24 hours plus a capstone agent certified in Attest to depth 2 or higher. In a manufacturer, that suits the operations engineers who configure planning and procurement agents. Shop-floor supervisors are better served by the Operator track. Our guide to AI literacy and certification for people who work with agents explains the levels.
Questions people ask
Is a predictive maintenance agent high-risk?
Usually not on its own. If it is a safety component of a product covered by EU product safety law, it may be; check the exact use and product with counsel.
Do shop-floor operators need AI literacy training?
If they use or are directed by agents, they need enough to understand what the agent does and when to raise a concern. That can be brief and practical.
Who should train first in a manufacturer?
Procurement and payments staff, because supplier fraud through agents causes immediate losses, and supervisors of allocation agents, because that use can be high-risk.