Jackson Lewis Publishes EU AI Act Guidance for Employers
Jackson Lewis has published guidance on assessing recruitment and HR tools under the amended EU AI Act, covering system classification, vendor contracts, oversight and bias checks. The November 17, 2026, podcast features Luís Neto Galvão of SRS Legal and Éric Le Quellenec of Flichy Grangé Avocats.
Employers should start with an inventory drawing on personal-data processing records, according to the guidance, and establish which tools use AI and for what tasks. That review should cover systems introduced locally, not just those approved centrally. For recruitment tools, the guidance recommends documenting their purpose and controlling the use of general-purpose chatbots with candidate information.
Under Annex III of the AI Act, high-risk employment categories encompass targeted recruitment advertising, application screening, candidate assessment and certain workplace decisions or monitoring. The Jackson Lewis guidance says arranging or storing CV information may carry a different classification from scoring, ranking or shortlisting applicants. It describes a limited exception for specified preparatory or procedural functions, but says that exception does not apply to systems that profile people.
According to EUR-Lex, Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, amended the Act and is in force. The Jackson Lewis guidance says the amendment shifted the application date for Annex III high-risk systems, including employment applications, to December 2, 2027. It gives August 2, 2028, as the later date for certain high-risk AI incorporated into regulated products.
The guidance also addresses whether an employer is a system’s deployer or provider. It says provider responsibilities may apply when an employer substantially alters a tool, puts its own name on it or changes its purpose in a way that makes it high-risk. Recommended vendor checks cover conformity documentation, while contract provisions should address instructions, bias information, logs, cooperation over incidents and system updates.
For high-risk deployments, the guidance sets out controls covering adherence to provider instructions, checks on employer-controlled input data, ongoing monitoring and log retention. Oversight should be assigned to people with authority to question or override outputs. It also recommends a documented approach to stopping a noncompliant or discriminatory tool and restoring a human process.
Checks for disparities between groups should examine actual recruitment, promotion and performance outcomes rather than vendor assurances alone, according to the guidance. Corrective action should be recorded, with testing repeated after substantial changes to the system or workplace. The guidance says processing special-category data for detecting and correcting bias is permitted only where strictly necessary, where anonymized or synthetic alternatives are insufficient, and where protections including pseudonymization are applied.
On staff literacy, the guidance says the amended obligation requires measures that support understanding of AI, rather than a guarantee of a particular proficiency level. It recommends role-specific training and records demonstrating that training took place. Personnel overseeing high-risk systems still require appropriate competence, training and authority, according to the guidance.
* This article is based on publicly available information at the time of writing.