Examen — Signing expert — the core track — Académie

Article 10 and the anatomy of a file

Q1

A dataset is purchased from a data broker. What's the first reflex under art. 10?

Use it: it was paid for, so it's legal Check the provenance and rights/licenses (ph1) and the legal basis for the use (ph2) Only check the available volume Nothing to check, that's the broker's problem

Q2

Complying with the applicable harmonized standards gives the provider:

A full exemption from the AI Act A presumption of conformity for the corresponding requirements A mere marketing label with no legal weight No regulatory benefit

Q3

What does "complete to the extent possible" mean under art. 10 for a dataset?

Containing all existing data on the subject Covering the cases and sub-populations relevant to the purpose, with no major gap Having no missing values whatsoever Being as large as possible

Q4

A provider states: "our model reaches 95% accuracy, so our data must be compliant." Where is the flaw in this reasoning under article 10? 2-3 sentences. Réponse libre — quelques phrases.

GDPR, sensitive data, and the §5 door

Q5

A provider proposes sending sensitive data to its cloud in the United States to "speed up the bias computation." What do you say? 2-3 sentences. Réponse libre — quelques phrases.

Q6

Once the model is trained, must the training dataset be kept indefinitely?

Yes, to enable future re-training No: retention is limited to the purpose, and any retention must be justified Yes, the law requires permanent retention for audit purposes It doesn't matter, training data is not subject to retention limits

Q7

Explain in 2-3 sentences the paradox that article 10 §5 resolves, and the main condition it imposes in return. Réponse libre — quelques phrases.

Q8

An HR dataset contains the "ethnicity" variable solely to measure a hiring bias. On what basis is this permissible?

The explicit consent of each individual, collected after the fact The controller's legitimate interest, with no other condition AI Act art. 10 §5: strictly necessary to detect the bias, surrounded by safeguards Prior anonymization, which takes the data outside the GDPR

Bias and ratios: reading a number without being fooled

Q9

Why is removing sensitive attributes from a dataset not enough to guarantee the absence of bias? 2-3 sentences. Réponse libre — quelques phrases.

Q10

The report shows a disparity ratio of 1.3 on loan approvals, between two comparably sized, robust groups. The client asks: "is that serious?" How do you answer as an expert? 2-3 sentences. Réponse libre — quelques phrases.

Q11

The "sex" variable was removed from the dataset, but the model remains biased by sex. Most likely explanation?

It's impossible without the variable: there must be a measurement error A proxy variable (e.g., job title, part-time status) indirectly re-encodes sex Random chance: rerunning it will make the gap disappear The model "memorized" the removed variable

Q12

Can all definitions of fairness be satisfied simultaneously?

Yes, you just need to optimize long enough No: several notions are mathematically incompatible as soon as base rates differ — one must choose and justify Yes, with enough data The question doesn't make sense

Judgment: sign or send back

Q13

Why is traceability (who signed what, on which version, when) as important as the content of the file? 2-3 sentences. Réponse libre — quelques phrases.

Q14

A well-crafted referral also serves to:

Show the client they were wrong Preserve the relationship by charting a clear path toward compliance Maximize billable time Nothing in particular

Q15

Is the expert's signature a guarantee that the system will never cause harm?

Yes, it guarantees the absence of any risk No: it attests the data's compliance with art. 10 at a given moment, not an outcome Yes, if the client countersigns it No, it has no value at all

Q16

What does a binding signature carry? (multiple answers) Plusieurs réponses possibles.

The signatory's identity The exact scope attested (sources, version) A fingerprint / seal guaranteeing integrity A promise of commercial results for the client

The professional framework: knowing and holding it

Q17

Before signing your first client file, what should reasonably be in place? (multiple answers) Plusieurs réponses possibles.

Active PI insurance At least one certification covering the AI Act / GDPR / audit pillars Mastery of the internal standard (the 8 sections and the bar) An exclusivity agreement with the client

Q18

A friend wants to become a signatory but is starting from zero (no law, no data, no compliance background). Where should they start, and what pitfall should they avoid? 2-3 sentences. Réponse libre — quelques phrases.

Q19

What makes up a credible professional setup for signing? (multiple answers) Plusieurs réponses possibles.

Active PI insurance At least one relevant certification Mastery of the internal standard (the 8 sections) A grand-sounding title one awards oneself

Q20

How are certifications like AIGP or CIPP/E kept current?

They are acquired for life, with no obligation Through periodic continuing professional education (CPE) credits By retaking the exam every month Simply by paying a membership fee, with no activity