Leveraging AI for Public Labor Intermediation in Spain: Bridging the Gap between Law 3/2023 and Administrative Practice in Labora ()
1. The “Modernization of Active Labor Market Policies” in Spain’S Recovery, Transformation, and Resilience Plan. The 2023 Employment Law
Reform 5, “Modernization of active labor market policies”, stands as a key element within Component 23 of Spain’s Recovery, Transformation and Resilience Plan. Implemented through Employment Law 3/2023 (hereinafter, LE-2023), this reform constitutes an ambitious reconfiguration of the Spanish labor market, with the strategic aim of transforming employment structures to reduce structural unemployment and improve the effectiveness of public policies.
The Preamble to LE-2023 articulated this objective by noting that our country needs to increase the capacity of our labor institutions in reskilling, guidance, and support for jobseekers, citing as a necessary measure the reform of active labor market policies, “a reform that must be based on measures such as the modernization of the Single Employment Portal through the use of artificial intelligence and other tools to improve the effectiveness and efficiency of matching labor supply and demand”. LE-2023 makes clear from its opening lines that the implementation of AI tools contributes to improving labor intermediation, which is part of the services included under Axis 1 of the six axes through which the active labor market policies of the annual employment plans are structured (currently, the Annual Plan for the Promotion of Decent Employment).
The 2023 Employment Act spans 70 pages in the Official State Gazette (PDF format). It comprises seventy (70) articles, eleven (11) Additional Provisions, five (5) transitional provisions, and sixteen (16) final provisions that amend various regulations; with respect to employment, these include Royal Decree 818/2021, September 28, 2021, which regulates the common employment activation programs of the National Employment System (Final Provision 6) [without altering its regulatory rank, Final Provision 13], and Royal Decree-Law 32/2021, December 28, 2021 on urgent measures for labor market reform (Final Provision 7). It also provides for the modernisation of the Single Employment Portal and improved coordination and cooperation with employment and training platforms (Final Provision 11), as well as provisions on the financing of guaranteed services (Final Provision 12). Throughout the statutory text, there are further references to the functions that AI can perform to enhance the aforementioned effectiveness and efficiency, ranging from Article 17—which constitutes the central reference point for legitimising decisions grounded in data analysis—to others that may appear less significant but are essential for the functioning of artificial intelligence, including, inter alia, institutional collaboration (Art. 48) and the use of these technologies to assist employment services in assigning to jobseekers the various services included in the Catalogue of guaranteed services (Arts. 55 et seq.). These guaranteed services include “efficient labor intermediation, which facilitates suitable job offers (…)” [Art. 56(f)]. In sum, this is a notable legislative effort, further complemented by the implementing regulation of Title IV through Royal Decree 438/2024, April 30, 2024, which develops the Common Service Portfolio of the National Employment System and the guaranteed services established in Law 3/2023, February 28, 2023, on Employment.
Developing an AI agent to improve public employment intermediation based on Deep Learning algorithms (Deep Learning) appears to be an essential and urgent task. Applying artificial intelligence to enhance the effectiveness and efficiency of Public Employment Services (PES) should have been their managers’ top priority, especially with regard to job placement. If “employment intermediation” refers to the activities of identifying and sourcing job vacancies, selection, matching, placement, and re-placement of workers, with the aim of providing workers with employment suited to their characteristics and facilitating employers’ access to workers best suited to their needs (art. 3), AI can provide substantial support by automating one or more of the processes required to carry out these activities. Nevertheless, some scholars (Fernández García, 2026) contend that the Common Service Portfolio currently restricts these capabilities to individualized tutoring and follow-up for job seekers, notwithstanding industry surveys indicating that their broader application would yield significant benefits for the alignment of labor supply and demand. It is necessary to remove the obstacles that prevent Public Employment Services from improving service delivery by leveraging the possibilities arising from algorithmic management of employment intermediation, while respecting workers’ rights (especially their fundamental rights) within the framework of the applicable legislation and, in particular, within the safeguards for high-risk systems established by Regulation 2024/1689, June 13, 2024 laying down harmonised rules on artificial intelligence. The notion that deep learning is inherently inexplicable due to its operational complexity is fundamentally flawed. These artificial intelligence systems are indeed amenable to explainability; furthermore, compared to simpler models, they offer the distinct advantage of yielding superior predictive performance.
Nearly two years after its enactment, the operational reality remains starkly different. Public administrations have struggled to pivot toward the technological shifts necessary for meaningful social and economic impact. This failure to adapt affects all unemployed individuals, but especially those whose personal circumstances make labor-market integration more difficult: the long-term unemployed, workers with disabilities, people over fifty, victims of gender-based violence, and young people. Alongside technical difficulties, a wide range of obstacles also arise, including a lack of political will and difficulties in achieving social and legal consensus, among others.
The actual use of these tools by private employment agencies is also not precisely known; it can be assumed that, to a greater or lesser extent, private employment agencies are using them at some stage of the intermediation process: résumé screening, candidate selection, and matching labor supply and demand. According to the Integrated Public Information System of Employment Services (SISPE), there are a total of 1799 private employment agencies, of which 389 operate exclusively through electronic means1. It should be noted from the outset that the implementing regulation governing employment agencies (Royal Decree 1796/2010, December 30, 2010) contains no reference whatsoever to the use of artificial intelligence algorithms in their labor intermediation activities and has not been reformed or amended to regulate their implementation in any way, thereby aligning its content with this recent development; however, the exclusive use of automated means to carry out intermediation activities results in their exclusion from the concept of labor intermediation (art. 3 LE-2023)2, which in turn raises other interesting legal issues.
This study employs a methodology that integrates primary documentary research with qualitative administrative analysis. The research is grounded in a comprehensive examination of LABORA’s internal documentation, which delineates the current procedures for labor intermediation. This analysis was further enriched by meetings and consultations with caseworkers and managers within the Valencian Employment Service to gain insights into the system’s operational reality. The legal-formal framework is built upon an analysis of key regulatory instruments—specifically Spain’s 2023 Employment Act and the EU Artificial Intelligence Act—alongside landmark jurisprudence from the Spanish Constitutional Court and the Court of Justice of the European Union. Rather than centering on a traditional review of academic doctrine, the objective of this paper is to scrutinize actual administrative practice as the necessary empirical starting point for designing AI-driven enhancements for public labor intermediation.
2. The Strategic Pillars of Le-2023
One of the most significant institutional changes introduced by LE-2023 was the transformation of the SEPE into the Spanish Employment Agency, with the aim of providing the central state body responsible for employment policy with greater functional and managerial autonomy in order to overcome the rigidity of the previous model. Title IV introduced a Common Portfolio of Guaranteed Services, granting its recipients (jobseekers, individuals, and firms) enforceable rights vis-à-vis the public employment services and moving beyond the former welfare-based model. These guaranteed services are as follows:
For job seekers (Art. 56): 1) Development of an individualized profile and design of a personalized pathway to improve employability; 2) individual tutoring and ongoing advice from a professional who guides and supports the job search; work-based training aligned with the individual’s profile and labor-market needs to enhance skills and professional qualifications.
For employers (Art. 57): 1) Management of submitted job vacancies, ensuring their dissemination and efficient processing; 2) Information and advice on recruitment and available support measures, facilitating access to incentives; 3) Identification of companies’ professional profile needs in order to align training provision and job intermediation accordingly.
The regulatory development of Title IV of LE-2023, implemented by the already cited Royal Decree 438/2024, April 30, 2024, also expressly provides for the large-scale use of data for the purpose of producing an individualized profile of the user of the system (Art. 12). This mixed profiling system (Art. 36 LE-2023) results in an assessment of the employability of the jobseeker. The concept of “employability” lies at the core of its strategy, defining it as both a “right and a duty” for jobseekers. Employability is defined as the set of transferable competences and qualifications that strengthen individuals’ capacity to take advantage of employment opportunities. This right/duty duality therefore means that jobseekers have a right to effective labor-market insertion through the public employment services and, in parallel, are obliged to contribute to that insertion by undertaking training actions and accepting offers of “suitable employment”. To ensure that the system is inclusive and reaches all workers, priority target groups are identified (Art. 50 LE-2023), including young people, the long-term unemployed, persons with disabilities, LGTBI persons (particularly trans people), those over 45 years of age, women who are victims of gender-based violence, persons experiencing social exclusion, and migrants.
3. Artificial Intelligence and Data Protection: Some Issues
3.1. The Regulatory Framework for High-Risk AI Systems under Spanish and EU Law
LE-2023 expressly legitimizes decisions based on data analysis (Art. 17) and establishes the SISPE (Integrated Public Information System) as the technical instrument for the recording of common data and the integration of information related to the management of active employment policies. the SISPE is the common information network of all public and private employment services, and LE-2023 guarantees the possibility of access to and large-scale processing of its data “for the purpose of profiling users of public employment services, as well as monitoring, evaluating the outcomes of active employment policies, and managing the guaranteed services regulated by this law”. Furthermore, LE-2023 (Art. 15): a) assigns to the SISPE the responsibility of ensuring that labor intermediation functions are carried out appropriately; b) requires the SISPE to be oriented toward continuous improvement in the quality, effectiveness, and efficiency of data; and c) provides that the National Employment System must allocate adequate human, financial, and material resources for the continuous optimization of the collection, management, and processing of data affecting employment and employability among applicants for public employment services and the satisfaction of labor demand. The information collected in the SISPE is essential because artificial intelligence agents rely on large-scale data processing, and it is crucial that the data used at all stages of these tools’ operation be adequate, both quantitatively and qualitatively.
With regard to data protection, Regulation (EU) 2024/1689 classifies AI systems used for the recruitment or selection of natural persons as high-risk systems. The processing of job applicants’ personal data by AI algorithms receives specific attention in the legal instrument (Art. 17). It should be recalled that, at the time LE-2023 was adopted, that EU instrument had not yet been adopted, which explains the absence of any reference to it in LE-2023 but, obviously, does not preclude its application. Regulation (EU) 2024/1689 characterizes AI systems used for the recruitment or selection of natural persons and, in particular, for “(…) analyzing and filtering job applications and evaluating candidates as a high-risk system” (Annex III). Indeed, it suffices to note that the safeguards provided for in Art. 17 LE-2023 are aligned with those established by the aforementioned EU instrument the principle of human oversight avoidance of direct and indirect discrimination, periodic assessment of their effects, algorithmic transparency and traceability, and protection of particularly sensitive data (Muñoz Ruiz, 2026). It will be necessary to await the regulatory development of the algorithmic instructions referred to in Art. 17.4 LE-2023 to determine how the requirements that the EU AI Regulation imposes on high-risk algorithms are integrated: quality management, technical documentation, transparency, and bias mitigation. As Rodríguez-Piñero Royo (2024) has observed, the enactment of the EU AI Act solidifies a burgeoning ‘Algorithmic Labor Law’ framework. While transversal in its application, this framework imposes stringent transparency and human oversight mandates upon recruitment and selection processes, reflecting their classification as high-risk systems due to their capacity to profoundly impact employment prospects and perpetuate historical biases. Furthermore, as Muñoz Ruiz (2026) emphasizes, the mandate for algorithmic transparency has transcended mere individual information rights to become a cornerstone of collective labor protections. This evolution is underscored by recent European jurisprudence—most notably the Court of Justice’s ruling of February 27, 2025 (Case Dun & Bradstreet, C-203/22)—which recognizes the right of affected parties to receive meaningful information regarding the underlying logic of automated processing. Within the context of public employment services, such transparency is vital not only for individual jobseekers but also for ensuring the robust human oversight and collective accountability required under Article 26 of the EU AI Act. Complementing these individual safeguards, Pérez del Prado (2026) argues that the integration of AI in labor markets requires a ‘networked regulatory’ approach, where worker representation plays a pivotal role. Specifically, the effective exercise of information and consultation rights serves as a vital collective check, ensuring that algorithmic systems are not only transparent to the individual but also subject to social dialogue and accountability within the broader institutional framework. n this context, Mercader Uguina (2026) posits that explainability and human oversight form the ‘backbone’ of fundamental rights protection in the algorithmic era. For high-risk systems such as labor intermediation, transparency must transcend general descriptions to provide ‘local’ or individualized explanations. Grounded in the counterfactual criteria validated by the Court of Justice (Case Dun & Bradstreet, C-203/22), such explanations must clarify not only the factors behind a specific outcome but also the conditions under which that outcome would have differed. Furthermore, effective human oversight is only possible if it addresses ‘automation bias’, ensuring that the public caseworker’s role is substantive and authoritative rather than a mere formal ratification of algorithmic outputs.
In any event, LE-2023 distinguishes between the general legal regime governing the processing of personal data (Art. 16) and the processing of data for the use of analytical tools and algorithm-assisted decision-making (Art. 17). Art. 16 operates as an enabling clause and a safeguards framework, ensuring that any processing of personal data by the Spanish Employment Agency (AEE) and the Public Employment Services (SPE) is subject to EU and national data protection rules (the GDPR and Organic Law 3/2018). In turn, Art. 17 focuses on specific safeguards for the processing of personal data by automated decision-making systems.
There are substantial differences between the two provisions and, for present purposes, particularly with regard to the processing of special categories of personal data3, the processing of which by algorithmic systems is completely prohibited; such processing is, however, permitted under Article 16, on the basis of the authorisation provided for in Article 9.2(b) GDPR4. Subject to this proviso, the personal data that may be processed are the same in both cases: “(...) those necessary and indispensable for the development and execution of the activities, techniques, and procedures established to guarantee the implementation of employment policy and (...) in particular, the identifying data of users of the services provided by the Spanish Employment Agency, the public employment services of the Autonomous Communities, and the public and private collaborating entities of the foregoing, as well as data relating to employment status and activity, education and training, social protection, and socioeconomic situation, among others” and their processing must be limited to what is indispensable (the principle of minimisation). Although these same data may be used for automated decision-making (“the information contained in SISPE”), the outcome must always be reviewed or modified by the personnel responsible for adopting the final decision.
3.2. The Paradox of Sensitive Data: Legal Barriers to Effective
Profiling
Identifying which data are indispensable (sufficient and appropriate) is essential. Some prioritized groups are prioritized precisely because of personal characteristics that are excluded from processing by the artificial intelligence agent: members of the Roma ethnic group, individuals in the LGBTI community, and even disability (as it may be classified as health data) or the status of victim of gender-based violence.
The disability status of job-seeking workers is subject to specific treatment. LE-2023 itself amended Article 38.2 of Royal Legislative Decree 1/2013, November 29, 2013, which approved the General Law on the Rights of Persons with Disabilities (hereinafter, LGDPD), by providing for the possibility of integrating into SISPE, subject to the prior consent of these workers, a reference to the type and degree of their disability. Indeed, Royal Decree 438/2024, April 30, 2024 suggests that registration of a job seeker as a person with a disability is voluntary. It should be recalled that the provisions of Organic Law 3/2018, December 4, 2018on the protection of personal data and the guarantee of digital rights are stricter than EU rules and preclude the mere consent of the data subject from lifting the prohibition on processing such “special” personal data; thus, this reform makes it possible to “circumvent” the provisions of the LOPD-2018, albeit while requiring the data subject’s consent. Therefore, on the basis of the GDPR, which “rehabilitates” the consent of the data subject as a legal basis for data processing, the type and degree of disability of the job seeker may appear among SISPE data; however, it is unclear whether this also extends to processing by the automated system. In my view, the answer should be affirmative.
In any event, this special rule does not resolve all the problems that the use of these particularly sensitive data may raise. These issues will not be addressed here, but it should suffice to note that the scheme chosen by the legislature for the processing of disability data appears to preclude the AI agent’s automated access to data contained in the registers or databases of other public administrations (especially the Tax Agency and the Social Security Administration). This greatly hinders the management capacity of such automated tools intended to support job-placement intermediation, which aligns poorly with the institutional cooperation and the effectiveness and efficiency advocated by LE-2023. Royal Decree 438/2024 provides (Art. 40) that public employment services are guaranteed access to, and the capability for mass processing of, SISPE data (which must be connected to the personalized employment file under Art. 42 of the same regulation) in order to carry out the profiling of users of intermediation services (among other purposes). With respect to the aforementioned single personalized employment file—which contains comprehensive information on the job-placement intermediation processes of the job seeker—the regulation expressly provides for the transmission of information from other administrations to the State Public Employment Service for the fulfillment of its statutory purposes, with specific reference to information relating to unemployment protection.
It appears that the implementing regulations for the instructions that will constitute the algorithm underlying the AI agent have not yet been adopted (Art. 17.4 LE-2023).
4. Analysis of the Administrative Procedure of the Valencian Employment Service (Labora)
4.1. Current Workflow for Private Vacancies and Employment
Programs
The workflow for posting job vacancies on the portal is similarly convoluted. After the company registers the vacancy in its application, it is assigned to a case manager, who reviews the information provided by the company (primarily to verify that it complies with the law and is non-discriminatory) and, based on this information and using the SPE’s internal IT system, completes the vacancy data. Once the case manager has finalized the registration, the vacancy is published on the Employment Portal, where it is available for jobseekers to consult and open to voluntary applications if they are interested. Jobseekers do not receive any notification or alert (SMS, email, or a notice in their profile on the Labora web portal) about the publication of a vacancy that could be aligned with their profile. The process concludes with the recording of the outcome in the database; this can be done by the recruiting company itself or, as is usually the case, by the PES case manager.
This term refers to programs developed through direct subsidies covering hiring costs. In some cases, these programs are aimed at recruitment by public entities, typically local authorities5. There are also employment promotion programs targeting private firms that serve “prioritised” groups.
Selection is normally carried out after a candidate scoring process. Its main feature in the case of hiring by local entities is its temporary nature, unlike hiring by private-sector employers. This temporariness—derived from the principle of merit and ability governing access to public employment—hampers the genuine labor-market integration of these workers—defined by the framework regulation as “unemployed persons registered as job seekers with Labora”—because, although it improves their employability, it exposes them to greater precariousness than other workers who benefit from legal reforms restricting fixed-term hiring. The subsidised items are wage costs and employers’ social security contributions. The target groups for these employment-promotion measures are specified in the various calls, and the Directorate-General of the Regional Employment Service is empowered to determine the selection criteria. These are the programs that specifically address “prioritised” groups.
In fiscal year 2025, the Valencian Community offered several employment promotion programs amounting to a total of €22,060,000 (€17,060,000 funded by the State Public Employment Service and €5,000,000 funded by the Generalitat Valenciana)6, aimed at supporting hiring by both local entities and private-sector employers. In the latter case, only open-ended, full-time hiring was subsidised (with a few very limited exceptions). These same calls are being repeated for 2026, although not all of the calls approved for 2026 have yet been approved.
The employment promotion programs approved during 2025 were as follows:
1) Program for the open-ended hiring, by employers in the ordinary labor market, of persons with disabilities coming from Special Employment Centers, and for the temporary hiring (between 3 and 12 months) of persons with severe disabilities7. Eligible beneficiary entities included private-sector employers, including self-employed individuals registered under the RETA. The 2026 call has already been published, with a maximum total budget allocation of €300,000—an amount identical to that planned for 2025—distributed as follows: Alicante: €122,700; Castellón: €32,670; Valencia: €144,6308.
2) Program for the open-ended hiring by private entities of unemployed persons belonging to specific priority target groups9. These groups were: a) individuals in, or at risk of, social exclusion, certified by the Social Services of any public administration; b) individuals experiencing long-term unemployment; c) persons over 50 years of age; d) persons with disabilities; and e) women who are victims of gender-based violence. Eligible beneficiaries also include private-sector employers, including individuals registered under the RETA. The same groups are prioritized in the 2026 call, with a total budget allocation of €4,500,000 distributed as follows: Alicante: €1,840,500; Castellón: €490,050; Valencia: €2,169,450. It should be noted that there is a substantial reduction in the budget allocated for 2026 compared with the amount ultimately devoted to this program in 2025, which was €11,760,000, and even compared with its initial allocation of €7,860,000, which was subsequently increased to the final amount indicated10.
3) Program for the permanent hiring of qualified young people within the National Youth Guarantee System11. The 2026 call includes a budget allocation with a maximum total amount of €3,700,000, distributed as follows: Alicante, €1,404,150; Castellón, €434,750; and Valencia, €1,861,100. This program has also seen a reduction in the total amount of funding in 2026 compared with 2025, when it amounted to €10,000,000 (Alicante: €3,690,000; Castellón: €1,162,000; Valencia: €5,148,000).
4) Recruitment program for unemployed individuals with severe mental disorders or severe mental health problems12. It is aimed at recruitment by local entities in the Valencian Community that have an active SASEM with the capacity to implement the program. The maximum program budget was €1,100,000.
5) Employment grant programs to support the hiring of unemployed individuals aged at least 30 years by local entities in the Comunitat Valenciana13. The maximum total amount for this program was €8,956,000.
6) Grant program aimed at hiring unemployed individuals by municipalities with fewer than 1000 inhabitants, or fewer than 5000 inhabitants located in counties at high or medium risk of depopulation, to carry out actions envisaged in emergency procedure plans in the forest domain14. The maximum overall amount was €18,500,000.
7) Hiring unemployed individuals under thirty years of age who are beneficiaries of the National Youth Guarantee System by associations of municipalities in the Valencian Community15. The maximum total amount is €3,879,000.
The selection criteria for workers are specified in the Resolution to be issued by the Directorate-General within the framework of each call, which is processed under a competitive procedure. The initial and sole criterion is the date on which the application is submitted, specifying that it includes both the application form and the remaining required documentation. If the date is the same, the time of submission is considered, and only if there is also a tie under this criterion are the following criteria applied: first, the disability status of unemployed individuals (in turn, prioritized according to the higher degree of disability); second, women; and, if the tie persists, the oldest application date, and, if this still does not resolve it, the lowest case-file number. A common feature of all these programs is that, although the beneficiary entity (local authorities) submits the job offer to the SPE, it is the latter that is responsible for preparing the prioritized list of selected individuals among registered unemployed persons who meet the established selection criteria.
4.2. Empirical Findings: Operational Bottlenecks and the Failure
of Reactive Matching
The analysis of the intermediation process conducted by the Valencian Employment and Training Service (LABORA) reveals a structural decoupling between legislative intent and administrative execution, which remains anchored in outdated procedures. The development of the internet and the subsequent creation of Employment Portals that allow certain administrative procedures to be carried out online were, at the time, a significant advance, but little further progress has been made over the past twenty years with regard specifically to the job intermediation process. The systemic obsolescence, rather than personnel performance, constitutes the primary barrier to effective intermediation through the Public Employment Service. This difficulty is compounded in the case of jobseekers belonging to the “priority” groups targeted by active labor market policies.
Before publication, each job posting undergoes a review and processing phase by the public caseworker, which consumes 80% of their time, because much of the information is collected in free-text fields that the caseworker must process (i.e., enter) manually. In addition, the jobseeker is responsible for preparing their CV and attaching it to the web portal application as a PDF file with no system-defined structure; therefore, the information it contains is not standardized. Moreover, updating the CV is also the jobseeker’s responsibility. Jobseekers frequently upload their CVs upon initial registration, only to let them languish for years without further updates. Updating, moreover, again requires the caseworker’s involvement, as they must manually enter the changes into the system after first verifying them.
There is also an “automatic matching” system, but it operates only on a subsidiary basis relative to the previous one, and its actual use appears to be very limited. For this matching, the system applies maximum caps (six candidates per vacancy), and the primary criterion it uses to rank the selected jobseekers is length of time since registration (as a jobseeker). In the management of private vacancies, membership in priority groups is not taken into account for any purpose, except at the company’s express request. Therefore, the prioritization of the groups repeatedly referred to in LE-2023 is nonexistent.
In addition, it should be noted that this is a reactive system, that is, it focuses exclusively on covering job vacancies. The labor intermediation process examined begins when the (private) company registers its vacancy in the IT application designed for this purpose. In other words, the regional PES does not undertake an active job search for jobseekers, not even for the unemployed, and even less for registered unemployed individuals belonging to any of the prioritized groups. Once the vacancy is published, it is unemployed individuals who are interested who voluntarily apply for the position offered. The nascent automated system that could match jobseekers based on the data available to the public employment service (Labora) is also underdeveloped and is not used in practice. Moreover, its design is also inadequate. The choice of an application-based system appears to reflect that hiring firms prefer to recruit genuinely interested individuals and view negatively the PES sending them workers whose registration as jobseekers is driven more by the desire to access different types of social benefits than by labor-market integration.
The process concludes with the recording of the outcome in the database; this can be done by the recruiting company itself or, as is usually the case, by the case manager. This process is also inefficient and likewise requires the public case manager to conduct individualized follow-up—typically by phone—on the vacancy.
This temporariness—derived from the principle of merit and ability governing access to public employment—hampers the genuine labor-market integration of these workers—defined by the framework regulation as “unemployed persons registered as job seekers with Labora”—because, although it improves their employability, it exposes them to greater precariousness than other workers who benefit from legal reforms restricting fixed-term hiring.
The foregoing reveals a disconnect between the regulatory objective of improving the effectiveness and efficiency of public employment services through the use of artificial intelligence tools as decision-support instruments and administrative practice, which remains entrenched in a “reactive” system in which the labor-market insertion of unemployed individuals is carried out primarily through the management of (private) job vacancies. The focus of job intermediation is placed on labor supply rather than demand. Moreover, applicants’ membership in the various prioritised groups is entirely irrelevant to the management of private job vacancies. In managing such vacancies, the public employment service neither conducts an active job search on behalf of jobseekers nor gives special consideration to jobseekers who may belong to any of the prioritised groups; indeed, in many cases it lacks this information. This lack of integration of active labor market policies into the intermediation of job vacancies is striking, as it contradicts both the text and the spirit of the Employment Act (2023) and its own programs promoting the hiring of prioritised groups.
5. Challenges and Proposals for Proactive Intermediation Based on Big Data Processing Tools
Furthermore, this cumbersome and sluggish administrative framework stands in stark contrast to the substantial public funding allocated for promote the hiring of individuals belonging to groups with particular difficulties in labor-market integration. Transforming the job-matching procedure as it is currently implemented is both essential and urgent. This transformation requires the immediate adoption of two measures.
First, it is necessary to apply the selection criteria for priority groups not only in employment promotion programs but also in the management of job vacancies (general or ordinary). Many micro and small enterprises, for various reasons (lack of awareness, bureaucracy, among others), do not request participation in public employment promotion programs for priority groups, thereby losing many opportunities for their labor market integration. Without this integration, the impact of active labor market policies on the most vulnerable groups will remain limited.
Along the same lines, it is noteworthy that public programs targeting priority groups exclude their hiring in the domestic work sector. The progressive equalisation of labor and social security rights, and even occupational risk prevention provisions, for domestic workers, together with the forthcoming change in the long-term care model, does not justify excluding this employment relationship from public employment-promotion programs. The long-term care sector constitutes a significant source of employment in ageing societies such as Spain. The transition towards a model that promotes deinstitutionalisation and the expansion of home-based care, as referred to in Component 22 of the Recovery, Transformation and Resilience Plan, cannot rely exclusively on public home-care services, which are not necessarily better. Direct hiring by the head of the household of a worker belonging to one of the priority groups to provide care for dependent persons entails adopting a broader (holistic) perspective, strengthening the human dimension on both sides of the employment relationship. It would also bring to light a substantial share of domestic work provided in the homes of dependent persons that, in many cases, is currently carried out under informal or unregulated conditions. The jurisprudence of both the Spanish Constitutional Court and the Court of Justice of the European Union clearly establishes the criteria necessary to prevent such affirmative action measures from constituting prohibited discrimination. According to the Constitutional Court, the existence of a factual situation of disadvantage justifies the adoption of compensatory measures, provided that they are reasonable, proportionate, and do not perpetuate gender stereotypes16. In addition to these requirements, the Court of Justice of the European Union mandates that such preference operate only in cases of equal merit, ensuring it does not apply automatically and allows for the assessment of the specific personal conditions and circumstances of the candidates17 ensuring that such measures remain a targeted tool for substantive equality rather than an immutable preference.
Second, it is essential to replace the current reactive system with proactive intermediation through the use of Deep Learning tools that conduct active job searches for jobseekers. This proactivity must encompass the entire service portfolio of the State Public Employment Service. Specifically, regarding labor market intermediation, it must be integrated into the processing of registrants’ curricula vitae, the management of job vacancies, and the implementation of automated matching systems designed to supersede the current applicant-driven model. Furthermore, this proactive approach should address the identification of training deficits and skills gaps within job seekers’ profiles, among numerous other applications. The persistent technological lag within the PES disproportionately burdens those facing the highest barriers to entry. All employment-promotion programs implemented by local entities to support the hiring of certain priority groups are temporary, and although such contracts may improve employability, they do not achieve effective labor-market insertion. Algorithmic management of intermediation, while respecting fundamental rights, has the potential to help remove these obstacles and to deliver a much more sophisticated matching (i.e., pairing of labor supply and demand) than the current approach, which is limited to occupation, professional level, or formal qualifications. Although the processing of sensitive data is complex due to the constraints imposed by data protection regulation (RGPD and LOPD), LE-2023 explicitly permits the processing, within SISPE, of data relating to the type and degree of disability, subject to the worker’s prior consent. The AI system could use this information—which is not currently among the default matching criteria used by the PES—to facilitate labor-market insertion. This possibility should also be extended to other data from which it can be inferred that the jobseeker belongs to any other prioritised group. In this way, the new intermediation system supported by AI agents could actively search for job vacancies that match the jobseeker’s profile. Since the final hiring decision rests with the posting firm, the recruitment of individuals belonging to priority groups could be encouraged by including, at a minimum, one jobseeker from these groups in the shortlist of selected candidates, even if the person does not meet all job requirements, accompanied by precise information on the economic advantages associated with their hiring. If the company ultimately hires the “prioritised” unemployed person, registering the contract with the PES should automatically trigger the process for applying for the established subsidies.
In addition, it could help simplify the administrative procedure for requesting these aids, which currently obliges the firm (and, to a lesser extent, the worker) to submit onerous administrative burden18. Evidently, there is a compelling opportunity to overhaul these procedures, such that the documentation required from the firm could be considerably reduced. This would transform the process by turning intermediation into a genuinely active tool for promoting hiring—i.e., one that focuses on unemployed individuals rather than exclusively on job vacancies—while also using direct marketing of subsidies as an incentive to hire workers with disabilities and, moreover, simplifying the procedure for applying for the established subsidies. In alignment with this call for modernization, Fernández García (2026) observes that while LE-2023 and the Common Service Portfolio recognize the digital dimension of employment services, they paradoxically restrict the use of AI tools to individualized tutoring and follow-up. He contends that such a narrow application overlooks the transformative potential of automated systems in the broader labor intermediation process, where AI could more effectively synchronize supply and demand on a national scale, thereby preventing public services from losing competitiveness against private digital intermediaries.
Ultimately, the intermediation system based on “mass data-processing tools” should:
Actively search for job openings compatible with the profile of the job seeker belonging to the prioritized groups.
Include at least one applicant from these groups in the shortlist of selected candidates (even if the applicant does not meet all the job requirements, provided that they meet the main or essential requirements).
Accompany the list of candidates with precise information on the economic advantages associated with hiring them.
Automate grant applications once the contract has been registered, drastically streamlining the procedure that currently requires up to seventeen documents to apply for financial support. This would transform intermediation into a tool for active promotion.
In any event, implementing any artificial intelligence system necessarily requires improving the data on job vacancies and jobseekers, since its proper functioning depends on their quantity and quality. It is essential to identify which data, at the current stage of digital development, are necessary/indispensable to achieve the objectives of the PES for matching, but also for career guidance and training; moreover, the artificial intelligence agent must have access to data concerning individuals’ membership in prioritized groups. In addition, it has been found that a large amount of data is requested—much of it unnecessary—yet many of these data are not mandatory, resulting in substantial disparities in the available information and adversely affecting matching quality. Implementing any AI system urgently requires, as a first step, transforming unstructured data (from vacancies and jobseeker profiles) into data of adequate quantity and quality; and the system’s design should allow algorithms to process vacancies and jobseeker profiles directly, without the public caseworker’s intervention except, to the minimum indispensable extent in the initial phases, in order to focus on final oversight of the outcome in accordance with Regulation 2024/1689, June 13, 2024, laying down harmonized rules on artificial intelligence. In this way, PES staff will be able to devote themselves to tasks with greater added value than the mere “data entry” of information included in text fields. Collaboration among public administrations should also be expanded by promoting interoperability and connectivity among their databases, as indicated by LE-2023.
Particular attention should be paid to the voluntariness governing whether jobseekers accept job offers. First, because the primary system for matching vacancies with jobseekers is structured around the candidate’s application to the job posting published on the Public Employment Service website. Moreover, once selected, it is up to the individual to accept or decline the job offer (telephone confirmation of availability). This voluntariness runs counter to both the letter and the spirit of the current regulatory framework, which requires beneficiaries to accept suitable employment through the signing of an activity agreement [arts. 256 c) and 299 c) and f) TRLGSS-2015 and art. 3.d) LE-2023], and even treats refusal of a job offer as a serious offence for recipients of unemployment benefits or assistance [art. 25.3 a) LISOS], punishable by loss or termination of benefits [art. 47.1b) LISOS]—sanctions that, for various reasons, appear not to be applied. It is unknown whether the regional Employment Service has information on whether the jobseeker/unemployed person is receiving unemployment benefits (contributory or assistance-based). If it lacks this information, that circumstance cannot be taken into account when assigning candidates to vacancies, which contradicts the linkage between active and passive labor market policies enshrined in the applicable legislation. Therefore, it is essential that the Public Employment Service have real-time access to this information. Where unemployed persons are not beneficiaries of benefits or assistance, refusing a suitable job offer entails a waste of resources, as the Public Employment Service has devoted considerable time to managing the vacancy or service offer. Free of charge does not mean cost-free. In a non-negligible (though unquantified) number of cases, some individuals register as jobseekers in order to access the benefits or advantages associated with that status but, for various reasons, do not wish to find employment. Therefore, solutions should be sought, not necessarily coercive, to prevent wasting resources on job-search efforts for individuals who do not want employment. For example:
allow registration with the public employment service by enabling a system under which the claimant is, temporarily and with justification, excluded from job search activities conducted by the public employment service. Royal Decree 438/2024, April 30, 2024, already provides for certain grounds for suspending registration as a jobseeker (Art. 9.2), but establishes that, by default, such suspension does not exclude the jobseeker from participation in job-matching processes. This provision is not particularly reasonable in some cases in which the suspension may be prolonged, such as deprivation of liberty, the care of children under 14 years of age or of persons with disabilities dependent on the jobseeker, or the care of relatives up to the second degree of consanguinity or affinity who are unable to care for themselves.
to enable a separate registration category for unemployed individuals who are not jobseekers, allowing access to social benefits that currently derive from registration as a jobseeker (similar to what occurred with the decoupling of the minimum wage and the IPREM), among other measures.
Ultimately, the Public Employment Service’s verification of a worker’s availability should be limited to confirming that the worker is not already employed and that the job offer is suitable, requiring the worker registered as unemployed to accept the job when it constitutes suitable employment. If the worker ultimately refuses it, their registration as a jobseeker could be suspended/terminated, adapting the provisions of unemployment regulations under Article 47.1(b) LISOS. Therefore, not only a change in procedures is required, but also the political will to apply the consequences provided by law in cases of unjustified refusal of job offers.
In this regard, the system for documenting the reasons why a job seeker rejects a job offer should also be improved. At present, there are significant differences depending on whether this information is recorded by the public employment service or by firms; the latter are very generic compared with the greater detail provided by the former. These reasons should be used to improve the job-matching process. Obviously, they can also be used to detect patterns of behavior among job seekers for monitoring, control, and, where appropriate, sanction. Registration as a job seeker constitutes a status treated as equivalent to being registered with social security for access to certain benefits, or serves as a condition for access to specific services or advantages in the public or private sphere. Given the complexity of the job-matching process and the amount of material and human resources it consumes, certain candidate behaviors (unjustified failure to attend the selection process; lack of interest in obtaining the job; withdrawal; unjustified non-acceptance) should be analyzed, especially when repeated. It would also be very useful, to improve the job-matching process, to know the reasons why a candidate is ultimately not hired by the offering firm; this would also feed into the public management pillars of career guidance and vocational training, which are also components of active labor market policies.
6. Conclusion
This study elucidates a significant disconnect between the ambitious legislative framework of the LE-2023 and the current administrative reality of public employment intermediation. While the law mandates the modernization of services through AI to enhance efficiency and equity, the case of LABORA reveals an institutional structure still anchored in reactive and manual procedures. Currently, caseworkers are burdened by “data entry” tasks that consume approximately 80% of their operational time, largely due to the prevalence of unstructured data and non-standardized curricula vitae. Furthermore, the findings indicate that “automatic matching” remains marginal and structurally flawed, as it prioritizes administrative seniority over the specialized needs of priority groups, who are often systematically excluded from standard job vacancies.
The core contribution of this work lies in the conceptualization and technical-legal roadmap for “proactive intermediation” supported by Deep Learning algorithms. This model departs from the traditional applicant-driven system to one where the AI agent actively searches for opportunities on behalf of the jobseeker. A pivotal advancement proposed herein is the mandatory inclusion of at least one candidate from a priority group in every shortlist, accompanied by the automated direct marketing of hiring incentives to employers. By integrating the promotion of subsidies directly into the matching process, the system transforms from a passive repository of vacancies into a dynamic tool for affirmative action that reduces administrative friction—potentially replacing the current 17-document manual application process with automated triggers upon contract registration.
Furthermore, this research contributes to the legal debate on data protection and high-risk AI by addressing the paradox of sensitive data processing. It argues that for public services to fulfill their constitutional mandate of effective labor insertion, the processing of disability or other “special category” data must be permitted under the strict safeguards of the EU AI Act and the GDPR, provided there is explicit consent and robust human oversight. Finally, this study advocates for an inclusive expansion of employment policies into the domestic work sector, aligning labor intermediation with the evolving long-term care model in aging societies.
In sum, this paper demonstrates that the transition to an AI-driven public employment service is not merely a technical upgrade but a legal and political necessity. By shifting from reactive management to proactive algorithmic intermediation, public administrations can finally bridge the gap between statutory guarantees and the effective labor-market integration of the most vulnerable citizens.
Funding
This work was conducted within the framework of the Research Project “Artificial intelligence for improving public employment intermediation” (PID2021-124978NB-I00), funded by the Ministry of Science and Innovation under the 2021 Call for Knowledge Generation Projects, for which the undersigned serves as Principal Investigator.
NOTES
1https://www.sistemanacionalempleo.es/AgenciasColocacion_WEB/listadoAgencias.do?modo=inicio (Date of last access: 13-2-2026).
2Art. 3.c) LE-2023: (…) In any event, for the described set of actions to be considered job intermediation or placement, they must not be carried out exclusively through automated means.
3Art. 9.1.: The processing of personal data revealing racial or ethnic origin, political opinions, religious or philosophical beliefs, or trade union membership, and the processing of genetic data, biometric data for the purpose of uniquely identifying a natural person, data concerning health, or data concerning a natural person’s sex life or sexual orientation, shall be prohibited.
4Art. 9(2)(b): “processing is necessary for the purposes of carrying out the obligations and exercising specific rights of the controller or of the data subject in the field of employment and social security and social protection law, insofar as it is authorised by Union or Member State law or a collective agreement pursuant to Member State law providing for appropriate safeguards for the fundamental rights and the interests of the data subject.”
5Order 13/2024, of 13 June, issued by the Regional Ministry of Education, Universities and Employment, which establishes the regulatory bases for subsidies for the hiring of unemployed persons by local entities in the Valencian Community, constitutes the foundational regulation underpinning the various calls for applications aimed at subsidizing hiring by local entities (DOGV of 18-6-2024).
See https://dogv.gva.es/datos/2024/06/18/pdf/2024_5764.pdf
6Resolution of 28 March 2025 of the Directorate-General, whereby subsidies are announced for fiscal year 2025 to promote the open-ended hiring of unemployed individuals belonging to certain priority groups for employment (DOGV of 3-4-2025). The initial budget allocation was increased by the Resolution of 28 August 2025 (DOGV of 4-9-2025). The RESOLUTION of 19 December 2025 of the Directorate-General, whereby subsidies are announced for fiscal year 2026 to promote the open-ended hiring of unemployed individuals belonging to certain priority groups for employment within the territorial scope of the Valencian Community (DOGVA DE 28-1-2026; https://dogv.gva.es/datos/2026/01/28/pdf/2026_2501_es.pdf), establishes the following budget allocation for this program: Alicante: €1,840,500; Castellón: €490,050; Valencia: €2,169,450.
7Resolution of 28 March 2025 of the Directorate-General, calling for applications for the 2025 financial year for grants to promote the temporary employment of persons with severe disabilities, as well as the transition from sheltered employment to the ordinary labor market, within the territorial scope of the Valencian Community (DOGV of 9-4-2025). Resolution of 14 July 2025 (DOGV of 17-7-2025) extended the deadline for submitting applications.
8For 2026, Resolution of 19 December 2025 of the Directorate-General, which announces, for fiscal year 2026, subsidies to promote the temporary hiring of persons with severe disabilities, as well as the transition from sheltered employment to the ordinary labor market, within the territorial scope of the Valencian Community (DOGV of 29-1-2026; https://dogv.gva.es/datos/2026/01/29/pdf/2026_2502_es.pdf)
9Decision of 28 March 2025 of the Directorate-General, announcing for fiscal year 2025 the grants aimed at promoting the indefinite hiring of unemployed persons belonging to specific priority groups for employment within the territorial scope of the Valencian Community (DOGV of 3-4-2025). Decision of 15 April 2025 of the Directorate-General, amending the Decision of 28 March 2025 (DOGV of 17-2025), extended the deadline for submitting applications. This deadline was extended a second time by the Decision of 24 April 2025 (DOGV of 28-4-2025). The deadline was amended for a third time by the Decision of 8 May 2025 (DOGV of 15-5-2025). It was amended for a fourth time by the Decision of 14 July 2025 (DOGVA DE 17-7-2025). The fifth extension of the deadline was carried out by the Decision of 3 September 2025 (DOGV DE 17-9-2025).
10The initial allocation for this program amounted to €7,860,000 (€7,060,000 from funds provided by the State Public Employment Service and €800,000 from the Generalitat Valenciana’s own funds). The Resolution of 28 August 2025 issued by the Directorate-General increased the overall amount by €3,900,000, bringing the total allocation up with funds from the Generalitat Valenciana (DOGV, 4 September 2025;
https://dogv.gva.es/datos/2025/09/04/pdf/2025_37482_es.pdf). The rationale for this increase was that the total amount of applications submitted across all provinces exceeded their respective allocations, and the Generalitat had available budget under the 2025 line for subsidies to promote the indefinite hiring of unemployed persons belonging to certain priority groups for employment within the territorial scope of the Valencian Community.
11Resolution of 28 March 2025, of the Directorate-General, convening for fiscal year 2025 the Programme to promote the open-ended employment of qualified young people, within the framework of the National Youth Guarantee System (DOGV of 7-4-2025;
https://dogv.gva.es/datos/2025/04/07/pdf/2025_9361_es.pdf)
12Resolution of 1 September 2025 of the Directorate-General, calling for applications for the 2025 financial year for subsidies aimed at hiring unemployed individuals with severe mental disorders or severe mental health problems by local entities in the Valencian Community (DOGVA of 5-9-2025; https://dogv.gva.es/datos/2025/09/05/pdf/2025_38226_es.pdf)
13Resolution of 7 July 2025 of the Directorate-General, calling for applications for the 2025 financial year for grants intended to support the hiring of unemployed persons aged at least thirty by local entities of the Valencian Community (DOGV of 10-7-2025;
https://dogv.gva.es/datos/2025/07/10/pdf/2025_26372_es.pdf)
14Resolution of 7 July 2025 of the Directorate-General, announcing, for the 2025 financial year, grants aimed at the hiring of unemployed persons by local entities of the Valencian Community to carry out actions envisaged in emergency plans or procedures in the forest domain (DOGV of 10-7-2025; https://dogv.gva.es/datos/2025/07/10/pdf/2025_26324_es.pdf)
15Resolution of 2 June 2025 of the Directorate-General, announcing for fiscal year 2025 the grants intended to support the hiring of unemployed persons under thirty years of age by associations of municipalities in the Valencian Community (DOGV of 5-6-2025).
See https://dogv.gva.es/datos/2025/06/05/pdf/2025_20065_es.pdf The maximum total amount is €3,879,000.
16See, inter alia, STC 128/1987, July 16; STC 19/1989, January 31; STC 269/1994, October 3; STC 229/1992, December 14.
17CJEU, October 17, 1995 (Kalanke, C-450/93); CJEU, March 28, 2000 (Badeck, C-158/97); and CJEU, July 6, 2000 (Abrahamsson, C-407/98).
18By way of example, the Resolution of 28 March 2025 of the Directorate-General, which announces for fiscal year 2025 the grants aimed at promoting the open-ended hiring of unemployed persons belonging to certain groups granted priority attention in employment policy within the territorial scope of the Valencian Community, lists up to seventeen documents that must accompany the application: a) bank direct-debit details; in the case of a new payee or a new bank account, the applicant must provide the account details through the dedicated online procedure entitled “Procedure to process registrations and cancellations of bank direct-debit details” (https://www.gva.es/es/inicio/procedimientos?id_proc=22648); b) a responsible declaration that the hiring does not fall under any ground for exclusion; c) a responsible declaration of compliance with the legislation on the labor-market integration of persons with disabilities or, where applicable, the exemption from that obligation; d) a responsible declaration regarding other aid received for the same eligible costs or the same subsidised event, or that none has been obtained; e) a responsible declaration that the applicant is not subject to the prohibitions on acquiring beneficiary status or on receiving payment referred to in paragraphs 2 and 3 of Article 13 and paragraph 5 of Article 34 of Law 38/2003, using the standard form; f) a responsible declaration of de minimis aid granted to the applicant person or entity during the three years prior to the date of submission of the grant application, using the standard form; g) notification to the hired person that their recruitment may be co-financed by the European Social Fund or by any other European Union fund under the aid requested, and information on the processing of their personal data, using the standard form; h) where applicable, a responsible declaration that the applicant is not subject to the prohibition on acquiring beneficiary status referred to in Article 13(3 bis) of Law 38/2003, using the standard form, together with, where applicable, the certification issued by an auditor registered in the Official Register of Auditors; i) a report from the General Treasury of the Social Security regarding the average workforce of workers registered as active during the thirty calendar days prior to the execution of the contract; j) a report from the General Treasury of the Social Security regarding the average workforce of workers registered as active in the period between the thirty days prior to hiring and the day of hiring of the hired person; k) where applicable, evidence of status as a woman victim of violence against women, pursuant to Article 9 of Law 7/2012 of 23 November of the Generalitat, Comprehensive Law against Violence against Women within the Valencian Community; l) proof that the hired person has completed the initial indicators form, in compliance with the obligations arising from Articles 16.1 and 42.1 and 42.2 of Regulation (EU) 2021/1060 laying down common provisions relating to certain European funds, including the European Social Fund Plus, and Article 17 and Annex I of Regulation (EU) 2021/1057 establishing the European Social Fund Plus (ESF+), both Regulations of the European Parliament and of the Council of 24 June 2021. (On LABORA’s website for the aid, the link to complete this form is available, and the proof generated upon completion is the document that must be submitted, signed by the hired person); m) the employment contract giving rise to the grant; n) notification to the Employment Service of the employment contract giving rise to the grant; o) where applicable, evidence of a situation of, or risk of, social exclusion issued by Social Services of any public administration; p) where applicable, a diagnosis of mental illness or mental disorder issued by the public mental health centre, in accordance with Article 2 of Order 12/2024; q) a responsible declaration that the applicant has been informed of the processing of personal data required for the management of these grants.