Sample: entirely fictitious example, not a client case.

Legal basis reviewed on 6 November 2026. Professional assessment, not legal advice.

1 System and role

System
AI visual inspection on the packaging line
Company
Mid-sized food manufacturer
Purpose
A camera with AI checks labels and closures and rejects faulty packs. People still decide on releasing the goods.
Role
The company operates the system. The plant manufacturer supplies the model and is the provider.
Boundary
Only the inspection station on this line. Release of goods and other lines are out of scope.

2 Classification with rationale

StepFindingRationale
AI systemYesThe model infers from camera images whether a pack is rejected.
RoleDeployerPurpose and model stay as supplied by the manufacturer. If that changes, the system is reclassified.
Prohibited practicesNoneThe application assesses packs, not people.
AI in regulated productsNot applicableThe AI has no safety function on the machine.
High-risk areas of useNone affectedThe inspection does not affect employees or decisions about people.
TransparencyNo specific obligationsThe AI does not talk to people and does not generate content.
Food safetyYesThe inspection is one of the controls in the HACCP plan. Changes go through change control.

Not a high-risk system under the EU AI Act. The AI literacy obligation under Art. 4 applies.

Not high-risk does not mean everything else is settled. That is why the pre-check of further areas of law follows right below.

3 Obligations

From the AI Act

  • Measures that support the development of the AI literacy of operators, shift leads and quality assurance (Art. 4 as amended by Regulation (EU) 2026/1744).

From the HACCP plan

  • A rule for what happens when the manufacturer updates the model. Every update is assessed before use and retested where needed.

Recommended

  • Spot checks of rejected parts to detect false alarms and missed defects.
  • A test record at commissioning with [detection rate per test record].

Dropped with a rationale

  • Everything the AI Act only requires for high-risk systems, because this application is not one. The rationale is recorded and reviewed with every change.

4 Pre-check of further areas of law

Ten questions per AI application, answered yes, no or unclear. They give the result for each area.

Area of lawResultNote
Data protectionopen, clarify firstWhether employees appear in the camera image, such as hands or faces, needs to be checked on site.
Works constitutionopen, clarify firstDepends on the same question. If people are in the image, the technology may be suitable for monitoring.
Trade secretsopen, clarify firstTo be clarified whether inspection images go to the manufacturer for model maintenance.
Equal treatmentnot affectedThe AI does not assess people.
Copyright, trade marks, unfair competitionnot affectedNo third-party works as input, no publication of results.
Personality rightsnot affectedThe AI does not show real persons, voices or names.
Cyber Resilience Act, product liabilitynot affectedThe AI is not part of a product that is sold.

Unclear answers remain open points until they are resolved. If an area is affected and the legal position is not clear, it goes to legal counsel.

5 Open points

  • Does the camera capture employees? Clarify with production management and data protection.
  • If so, inform the works council in good time and check co-determination.
  • Do inspection images go to the manufacturer? Check the contract and the manufacturer's statements.
  • [Detection rate per test record] is not yet available.

6 Next steps

  1. Resolve the open points and complete the pre-check.
  2. Define the AI literacy concept for operators, shift leads and quality assurance.
  3. Add a rule for model updates to change control.
  4. Test record at commissioning, then approval by production management.
  5. Review the classification for a new model, purpose or data, otherwise yearly.

Reservation

This sample is a professional assessment and not legal advice. The legal basis was reviewed on 6 November 2026. Obligations for high-risk systems under Annex III apply from 2 December 2027, under Annex I from 2 August 2028. Use of data from connected products, data intermediation, platform law and sector law are not part of the pre-check.

Sample: entirely fictitious example, not a client case.

What would this look like for your AI?

Book an initial call

30 minutes, confidential, no obligation