EU AI Act · recruitment & CV screening
Screen CVs with people. Not algorithms.
Toloka screens and shortlists candidates with vetted human specialists working from formal, de-identified criteria. No AI system makes the call — so the AI Act’s high-risk obligations never attach to your screening, and you keep the speed and cost of automation.
Screening decision record
de-identified
Warehouse Operative — Rotterdam DC
ref CV-48217 · personal data removed
Formal criteria
Result
C1
Valid EU forklift licence
Met
C2
≥ 12 months warehouse experience
Met
C3
Available for rotating shifts
Partial
C4
Right to work in NL
Met
C5
Manual handling ≥ 20 kg (declared)
Met
Decision:
Advance to shortlist
Reviewer TLK-R-3391 · confirmed by TLK-R-2740 · Logged 14:32 CET · criteria v2.1 · written rationale attached
Trusted by Leading AI Teams
The regulation
AI that screens candidates is high-risk under the Act
The EU AI Act classifies AI systems used to filter, rank or assess job applicants as high-risk. That places a substantial set of obligations on whoever deploys the tool — and a screening process run by people sits outside that regime entirely.
Human oversight architecture
Risk management system
Conformity assessment
Technical documentation
Annex III registration
Post-market monitoring
The process
Four steps, fully managed
A disciplined human pipeline that produces a defensible record for every candidate.
![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 48 48" xmlns="http://www.w3.org/2000/svg"><path d="M 29.45 22.3 L 27.25 20.1 C 28.117 17.733 27.667 15.767 25.9 14.2 C 24.133 12.633 22.217 12.233 20.15 13 L 17.95 10.8 C 18.517 10.433 19.15 10.167 19.85 10 C 20.55 9.833 21.267 9.75 22 9.75 C 24.367 9.75 26.375 10.575 28.025 12.225 C 29.675 13.875 30.5 15.883 30.5 18.25 C 30.5 18.983 30.408 19.708 30.225 20.425 C 30.042 21.142 29.783 21.767 29.45 22.3 Z M 35.9 28.75 L 33.9 26.75 C 35.533 25.55 36.958 24.208 38.175 22.725 C 39.392 21.242 40.283 19.75 40.85 18.25 C 39.183 14.55 36.683 11.625 33.35 9.475 C 30.017 7.325 26.4 6.25 22.5 6.25 C 21.1 6.25 19.667 6.383 18.2 6.65 C 16.733 6.917 15.583 7.233 14.75 7.6 L 12.45 5.25 C 13.617 4.717 15.108 4.25 16.925 3.85 C 18.742 3.45 20.517 3.25 22.25 3.25 C 27.017 3.25 31.375 4.608 35.325 7.325 C 39.275 10.042 42.167 13.683 44 18.25 C 43.133 20.383 42.017 22.333 40.65 24.1 C 39.283 25.867 37.7 27.417 35.9 28.75 Z M 38.8 40.05 L 30.4 31.8 C 29.233 32.267 27.917 32.625 26.45 32.875 C 24.983 33.125 23.5 33.25 22 33.25 C 17.133 33.25 12.717 31.892 8.75 29.175 C 4.783 26.458 1.867 22.817 0 18.25 C 0.667 16.517 1.592 14.825 2.775 13.175 C 3.958 11.525 5.4 9.95 7.1 8.45 L 0.8 2.15 L 2.9 0 L 40.75 37.85 Z M 9.15 10.55 C 7.917 11.45 6.725 12.633 5.575 14.1 C 4.425 15.567 3.6 16.95 3.1 18.25 C 4.8 21.95 7.358 24.875 10.775 27.025 C 14.192 29.175 18.067 30.25 22.4 30.25 C 23.5 30.25 24.583 30.183 25.65 30.05 C 26.717 29.917 27.517 29.717 28.05 29.45 L 24.85 26.25 C 24.483 26.417 24.033 26.542 23.5 26.625 C 22.967 26.708 22.467 26.75 22 26.75 C 19.667 26.75 17.667 25.933 16 24.3 C 14.333 22.667 13.5 20.65 13.5 18.25 C 13.5 17.75 13.542 17.25 13.625 16.75 C 13.708 16.25 13.833 15.8 14 15.4 Z M 24.4 17.65 M 18.6 20.55" fill="rgb(0, 0, 0)" height="40.050000000000004px" id="ohNMFBS5d" transform="translate(2 4.75)" width="43.99999885559083px"/></svg>)
01 — De-identify
Remove the person
Names, photos, age, gender, nationality and other identifiers are stripped from every CV before review — programmatically and consistently. Reviewers see qualifications, not people.
![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 48 48" xmlns="http://www.w3.org/2000/svg"><path d="M 24.8 31.9 L 22.7 29.8 L 28.25 24.25 L 22.7 18.7 L 24.8 16.6 L 30.35 22.15 L 35.9 16.6 L 38 18.7 L 32.45 24.25 L 38 29.8 L 35.9 31.9 L 30.35 26.35 Z M 28.95 13.2 L 21.85 6.1 L 23.95 4 L 28.95 8.95 L 37.9 0 L 40 2.15 Z M 0 25.4 L 0 22.4 L 18 22.4 L 18 25.4 Z M 0 9.4 L 0 6.4 L 18 6.4 L 18 9.4 Z" fill="rgb(0, 0, 0)" height="31.900006103515636px" id="ORK106hMP" transform="translate(4 8.1)" width="40.00000228881834px"/></svg>)
02 — Define
Formalise the role
We work with your team to decompose each role into objective, documented pass/fail criteria — the same standard applied to every applicant, and versioned over time.
![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 48 48" xmlns="http://www.w3.org/2000/svg"><path d="M 16 15 C 13.8 15 12 14.3 10.6 12.9 C 9.2 11.5 8.5 9.7 8.5 7.5 C 8.5 5.3 9.2 3.5 10.6 2.1 C 12 0.7 13.8 0 16 0 C 18.2 0 20 0.7 21.4 2.1 C 22.8 3.5 23.5 5.3 23.5 7.5 C 23.5 9.7 22.8 11.5 21.4 12.9 C 20 14.3 18.2 15 16 15 Z M 0 31.05 L 0 26.35 C 0 25.083 0.317 24 0.95 23.1 C 1.583 22.2 2.4 21.517 3.4 21.05 C 5.633 20.05 7.775 19.3 9.825 18.8 C 11.875 18.3 13.933 18.05 16 18.05 C 18.067 18.05 20.117 18.308 22.15 18.825 C 24.183 19.342 26.315 20.087 28.546 21.06 C 29.589 21.531 30.426 22.213 31.055 23.108 C 31.685 24.003 32 25.083 32 26.35 L 32 31.05 Z M 3 28.05 L 29 28.05 L 29 26.35 C 29 25.817 28.842 25.308 28.525 24.825 C 28.208 24.342 27.817 23.983 27.35 23.75 C 25.217 22.717 23.267 22.008 21.5 21.625 C 19.733 21.242 17.9 21.05 16 21.05 C 14.1 21.05 12.25 21.242 10.45 21.625 C 8.65 22.008 6.7 22.717 4.6 23.75 C 4.133 23.983 3.75 24.342 3.45 24.825 C 3.15 25.308 3 25.817 3 26.35 Z M 16 12 C 17.3 12 18.375 11.575 19.225 10.725 C 20.075 9.875 20.5 8.8 20.5 7.5 C 20.5 6.2 20.075 5.125 19.225 4.275 C 18.375 3.425 17.3 3 16 3 C 14.7 3 13.625 3.425 12.775 4.275 C 11.925 5.125 11.5 6.2 11.5 7.5 C 11.5 8.8 11.925 9.875 12.775 10.725 C 13.625 11.575 14.7 12 16 12 Z M 16 7.5 M 16 28.05" fill="rgb(0, 0, 0)" height="31.050003051757812px" id="kDXm91xaW" transform="translate(8 8.95)" width="32px"/></svg>)
03 — Review
People decide
Vetted specialists score each de-identified CV against the criteria. A human makes every decision; for sensitive roles, a second reviewer confirms it.
![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 48 48" xmlns="http://www.w3.org/2000/svg"><path d="M 5.25 40 C 3.792 40 2.552 39.49 1.532 38.468 C 0.511 37.448 0 36.208 0 34.75 L 0 28.5 L 6.35 28.5 L 6.35 0 L 9.34 3 L 12.33 0 L 15.32 3 L 18.31 0 L 21.3 3 L 24.3 0 L 27.3 3 L 30.3 0 L 33.3 3 L 36.3 0 L 36.3 34.75 C 36.3 36.208 35.79 37.448 34.769 38.468 C 33.748 39.49 32.508 40 31.05 40 Z M 31.05 37 C 31.717 37 32.258 36.792 32.675 36.375 C 33.092 35.958 33.3 35.417 33.3 34.75 L 33.3 5 L 9.35 5 L 9.35 28.5 L 28.8 28.5 L 28.8 34.75 C 28.8 35.417 29.008 35.958 29.425 36.375 C 29.842 36.792 30.383 37 31.05 37 Z M 12 12.9 L 12 9.9 L 24 9.9 L 24 12.9 Z M 12 19.6 L 12 16.6 L 24 16.6 L 24 19.6 Z M 28.65 12.9 C 28.25 12.9 27.9 12.75 27.6 12.45 C 27.3 12.15 27.15 11.8 27.15 11.4 C 27.15 11 27.3 10.65 27.6 10.35 C 27.9 10.05 28.25 9.9 28.65 9.9 C 29.05 9.9 29.4 10.05 29.7 10.35 C 30 10.65 30.15 11 30.15 11.4 C 30.15 11.8 30 12.15 29.7 12.45 C 29.4 12.75 29.05 12.9 28.65 12.9 Z M 28.65 19.35 C 28.25 19.35 27.9 19.2 27.6 18.9 C 27.3 18.6 27.15 18.25 27.15 17.85 C 27.15 17.45 27.3 17.1 27.6 16.8 C 27.9 16.5 28.25 16.35 28.65 16.35 C 29.05 16.35 29.4 16.5 29.7 16.8 C 30 17.1 30.15 17.45 30.15 17.85 C 30.15 18.25 30 18.6 29.7 18.9 C 29.4 19.2 29.05 19.35 28.65 19.35 Z M 5.2 37 L 25.8 37 L 25.8 31.5 L 3 31.5 L 3 34.75 C 3 35.417 3.211 35.958 3.633 36.375 C 4.054 36.792 4.577 37 5.2 37 Z M 3 37 L 3 31.5" fill="rgb(0, 0, 0)" height="40px" id="Fbs_1KvLP" transform="translate(5.85 4)" width="36.30000000000001px"/></svg>)
04 — Document
Log everything
Each decision is recorded with the reviewer, the criteria version and a written rationale — a complete, exportable audit trail per candidate.
Why legal teams choose this
Out of the high-risk perimeter, by design
What a human-first process removes
No conformity assessment for a screening tool
No Annex III high-risk registration for the activity
No Article 14 oversight architecture to build and evidence
No high-risk technical documentation and monitoring regime
What still applies — and how we help
GDPR applies to any hiring process; de-identification supports data minimisation
A documented human decision strengthens your position on automated-decision rules
Equal-treatment law applies; consistent criteria reduce disparate-treatment risk
Data handled under ISO 27701 and GDPR, with private or on-premises storage
Because no AI system screens or ranks candidates, the screening activity falls outside the AI Act’s high-risk requirements for recruitment. Your wider hiring process remains subject to data-protection and employment law — where this approach is built to strengthen, not weaken, your position.
Cost & speed
Manual review, automated economics
Human screening has a reputation for being slow and expensive. Ours isn’t. Roles are decomposed into discrete, well-specified review tasks and routed across a global specialist workforce, with automated quality control on every output and a tiered mix matched to each task’s difficulty. You pay only for work that passes QA.
A managed service, not a tool to staff
Toloka runs the entire pipeline — workforce sourcing and vetting, task design, quality control, reporting and delivery — to production SLAs. Your team approves the criteria and receives shortlists and records. Nothing to build, and no annotators to manage.
Security & data handling
ISO 27001
ISO 27701
ISO 9001
SOC 2 Type II
GDPR
CCPA
HIPAA
Candidate data is processed under audited controls, with private and on-premises storage options. Toloka AI BV is headquartered in Amsterdam.
Already using AI screening?
We can be your human oversight layer
If you keep an AI screening tool, the Act requires meaningful human oversight — people with the competence and authority to review, override and halt the system, and evidence that they do. Toloka supplies that review workforce and the documented oversight trail to support your Article 14 obligations.
Accountability stays with you as the deployer. We make oversight operable, consistent and evidenced — not a box-ticking exercise.
The data partner behind frontier AI teams
Quality-controlled, on-demand human data — delivered to production SLAs.
10+ yrs
Managing human work at scale
200,000+
Specialists
90+
Domains incl. law & medicine
EU-based
Headquartered in Amsterdam
Bring us a role you hire for at volume
We’ll show you the criteria model, the records you’d receive for each candidate, and the numbers — turnaround, quality and cost — for your hiring.