Level 5 exam — Generative AI at Work

Exam — fresh questions

12 new questions to validate Ethics, security & legal framework: confidentiality, checking, bias/ethics/IP, GDPR/AI Act, usage charter.
Answer everything, then click Check: your score and explanations will appear.

5.1 · Recall

1. What is the rule that beats all others at level 5?

Delete your conversations after each use Never entrust sensitive data to an unapproved tool Use as many tools as possible to compare Quickly anonymising everything is enough

Classify before pasting; if in doubt, do not paste — approved tool required.

5.1 · Application

2. Amine must summarise confidential minutes (names, amounts). Correct move?

Disable training then paste the raw text, that is enough Paste it raw into a public assistant to go fast Classify, anonymise for real (cross-checks tested) then approved tool Do nothing and give up using AI

Pseudonymisation alone insufficient; anti-training setting necessary but not sufficient outside an approved framework.

5.2 · Recall

3. What is the responsibility principle of checking?

Whoever signs checks: responsibility belongs to whoever releases Nobody is responsible for AI content “The AI said so” is enough as a defence The tool vendor is always responsible

Checking is proportionate to the stakes; figures, quotes and references, always.

5.2 · Application

4. Amine releases an AI report with a rate and a reference. Correct protocol?

Ask for sources, check against the source, hunt for contradiction Cross-check two models and stop if they agree Only check if the client asks a question Re-read the style and release, the source will follow

Level by stakes; two models agreeing is no substitute for checking against sources.

5.3 · Application

5. What should be done as soon as AI content touches people?

Release it as is, AI is neutral Imitate a known style to embellish it Assume the content is protectable without reading the ToS Anti-bias re-reading (genders, stereotypes, exclusions)

AI can neutralise its own biases — if explicitly asked.

5.3 · Application

6. Amine wants a visual “in the manner of a living illustrator” for commercial use. Decision?

Generate and release, it is inspiration Generate crediting the artist in small print Drop it: high-risk zone, licensed bank instead Generate then change two details to erase the copy

Style imitation of living artists ruled out commercially; ToS read for IP.

5.3 · Arbitration

7. Generated job ad that masculinises technical jobs, urgent release requested. Amine's decision?

Release fast, fix later if someone notices Neutralise the biases with AI, re-read, only then release Delete the ad and give up AI for good Release stating that AI wrote it

Mandatory anti-bias re-reading before release; the AI mention does not fix bias.

5.4 · Recall

8. Faced with personal data or a use touching people, which reflex?

Automate the whole decision to be efficient Self-interpret the texts among colleagues Wait for an incident to react Minimisation, human in the loop, question to the DPO/lawyer

Spot sensitive situations, do not settle them alone: the DPO/lawyer decides.

5.4 · Application

9. Amine wants AI to sort CVs. Correct move?

Paste the CVs into a public tool Let AI alone reject weak applications Score candidates in a spreadsheet + AI on the sly Do not improvise: minimisation, human decides, question to the DPO

Hiring = sensitive zone with demanding compliance; never improvise nor shadow AI.

5.4 · Arbitration

10. No approved tool in the company and pressure to use AI on internal data. Decision?

Ban all AI in the team to set an example Anonymise “quickly” and consider the matter settled Public AI only on public/generic + escalate the need for an approved tool Still use public tools on internal data

No approved tool does not authorise sensitive data in public tools; it must be escalated.

5.5 · Application

11. Amine writes the team charter. Correct content?

A PDF signed once then filed with no discussion One page, five sections, grey cases tested, approval + dated review Total ban of all AI in the team Total laissez-faire with an encouraging sentence

Neither ban (shadow AI) nor laissez-faire (incidents): the living middle way.

5.5 · Arbitration

12. The team already has a group AI policy. Is a team charter still needed?

Yes: the charter applies the policy in concrete moves (tools, cases, approval), then lives in meetings Yes: to replace and contradict group policy No, group policy always suffices in every respect No, team charters are forbidden in principle

The charter cites and applies the policy; approved (manager, IT/DPO, lawyer if in doubt), reviewed on date.

What's next?

Score ≥ 80%: move to Level 6 “In practice” — real cases and toolbox (NO exam).
Score 50-79%: re-read the explanations above — each re-explains the concept — then redo the 48h challenges of the failed sections.
Score < 50%: revisit sheets 5.1 (data) and 5.2 (checking), your two vital moves, before continuing.