Jackson Lewis podcast urges employer AI checks despite delayed EU deadlines
The lawyers distinguish basic résumé processing from candidate ranking, and argue that meaningful oversight requires authority to reject an AI recommendation.
Listen to this story
The audio brief
Story brief
3 key pointsIn a November 17, 2026, We get work® episode, Jackson Lewis lawyers advise employers to use the EU AI Act’s delayed deadlines to prepare—not to assume compliance can wait. The lawyers say the Digital Omnibus took effect July 27, 2026, and moved the deadline for listed high-risk systems, including recruitment tools, to December 2, 2027; they note that prohibitions and transparency rules already apply. For employers, the immediate work is to find how AI is actually being used in hiring and ensure people can meaningfully challenge or stop its outputs.
- 01
The lawyers distinguish résumé data extraction, which probably is not high risk, from scoring, ranking, or shortlisting candidates, which they classify as high-risk uses.
- 02
Workplace emotion recognition is prohibited except for medical or safety reasons; the episode gives video interviews scoring enthusiasm from facial expressions as an example.
- 03
Employers should inventory tools used by recruiters and local managers, including AI features added to existing HR software through updates.
Jackson Lewis published a November 17, 2026 podcast urging employers to prepare their hiring and workplace AI systems for compliance despite postponed EU deadlines. On We get work®, lawyers Luís Neto Galvão and Éric Le Quellenec explain why extra time for some high-risk systems does not put every obligation on hold.
In the episode transcript, Le Quellenec says the Digital Omnibus amendment took effect July 27, 2026. He places the revised deadline for listed high-risk systems, including recruitment and career management, at December 2, 2027. But he says prohibitions have applied since February 2025 and transparency rules since August 2026. Those dates frame the lawyers’ advice, rather than a newly announced regulatory change.
Sorting résumés is not the same as ranking people
The first step is identifying what a tool actually does. Le Quellenec distinguishes software that follows fixed rules from AI that infers predictions, recommendations or decisions from inputs. A spreadsheet filtering applicants by years of experience is not AI in his example; a system learning from past hiring decisions to rank applicants is.
- Preparatory work: Neto Galvão says extracting résumé data into an applicant database probably falls outside the high-risk category.
- Selection: The lawyers classify scoring, ranking and shortlisting candidates as high-risk uses, rather than mere administration.
- Prohibited use: Le Quellenec identifies workplace emotion recognition as prohibited except for medical or safety reasons, citing video interviews that score enthusiasm from facial expressions.
The inventory should extend beyond procurement records, Neto Galvão advises. Recruiters and local managers may adopt tools directly, while existing HR software can acquire AI functions through updates. He recommends starting with existing data-processing records under the General Data Protection Regulation, then checking what teams actually use.
Oversight needs an off switch
Their second step combines governance and vendor contracts. Le Quellenec says the vendor usually provides the system and the employer deploys it. However, an employer can assume heavier provider responsibilities by substantially modifying a tool or changing its purpose so it becomes high risk.
For human oversight, Neto Galvão recommends defining which decisions remain exclusively human, when recommendations must be challenged, and who can stop the tool. A manager restricted to confirming its output does not provide meaningful oversight, he argues. Le Quellenec also describes monitoring, incident reporting, worker notices and keeping logs for at least six months among deployer duties.
Contracts should secure information about bias testing, accuracy, logs, incidents and updates, Le Quellenec says. He also urges employers to prepare a human fallback if a system proves discriminatory or noncompliant. That turns the ability to stop using AI into a practical requirement for the workflow, not just an escalation policy.
Test outcomes and train for the actual job
For bias mitigation, the third step, Le Quellenec recommends comparing selection, promotion and performance outcomes between groups instead of accepting vendor assurances. Disparities should be investigated and corrective action documented. Testing should recur after significant updates or changes to the workforce or use case.
That testing can create a privacy tradeoff. Neto Galvão says the amended rules allow sensitive personal data to be used to detect and correct bias only when strictly necessary, and when synthetic or anonymized data will not suffice. He describes safeguards including pseudonymization, security measures and deletion once the bias is corrected.
Training is the fourth step. Le Quellenec says the amended literacy requirement asks companies to support staff understanding rather than guarantee a sufficient level. The lawyers recommend documented, role-based instruction: recruiters need to recognize ranking errors and discriminatory proxies, while procurement teams need to ask vendors the right questions.
Their fifth step brings those controls into recruitment itself. Le Quellenec warns that asking a general-purpose chatbot to rank résumés may turn it into a high-risk use and potentially make the employer a provider. Employers therefore need to control how tools are used, not merely which tools are approved.
Sources
- podcasts.apple.comFive Steps for Amended EU AI Act Compliance
- jacksonlewis.comFive Steps for Amended EU AI Act Compliance - Jackson Lewis
Reader comments
Newest comments first. Replies stay oldest first.