NEW TECHNOLOGIES AT THE WORKPL ACE COLLECTIVE BARGAINING PRACTICES ON AI AND ALGO­ RITHMIC MANAGEMENT IN EUROPEAN SERVICES SECTORS Simona Brunnerová, Daniela Ceccon, Barbora Holubová, Marta Kahancová, Katarína Lukáčová, Gabriele Medas This report was prepared in cooperation between UNI Europa and the FES Competence Centre on the Future of Work by the Central European Labour Studies Institute and Wage­ Indicator Foundation.

Contents EXECUTIVE SUMMARY 3 INTRODUCTION 4 1 CURRENT KNOWLEDGE ON AI IN COLLECTIVE BARGAINING AND COLLECTIVE AGREEMENTS 6 1.1 Examples from Italy and the UK ............................................................. 6 1.2 Example from Spain ............................................................................. 7 2 FINDINGS OF THE SURVEY AMONG UNI EUROPA AFFILIATES 8 2.1 The overall relevance of AI-related collective bargaining ............................. 8 2.2 Topics covered in AI-related collective bargaining ...................................... 8 2.3 Relevance of particular topics in AI-related collective bargaining .................. 11 2.4 Existing collective bargaining agreements with AI-related topics .................. 15 2.5 Reasons for the absence of AI-related provisions in collective agreements ..... 16 2.6 Other tools and measures related to AI ................................................... 16 3 ANNOTATION, CODING AND ANALYSIS OF COLLECTIVE AGREEMENTS’ CONTENTS 17 3.1 Analysis of CBAs .................................................................................. 18 3.2 AI-related content in CBAs .................................................................... 20 4 CONCLUSIONS AND LESSONS LEARNED 23 REFERENCES 24 ANNEXES 25 1. Sample of online survey responses among UNI Europa affiliates .................. 25 2. Additional findings from the online survey of the UNI Europa affiliates ......... 26 3. AI-related content in CBAs .................................................................... 28 1

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors LIST OF FIGURES Figure 1: Relevance of the AI-related collective bargaining............................................................................................. 9 Figure 2: Perceived impact of AI-related bargaining on the workforce, job quality and training...................................... 9 Figure 3: Perceived impact of AI-related bargaining on trade unions’ rights................................................................... 10 Figure 4: Opinion on the use of AI-related bargaining on the recruitment process, work organisation and workers’ assessments..................................................................................................................................................... 10 Figure 5: Opinion on the use of AI tools for worker surveillance in CB........................................................................... 11 Figure 6: Opinion on data protection in CB.................................................................................................................... 11 Figure 7: Bargained AI-related topics by trade unions..................................................................................................... 12 Figure 8: AI topics not bargained, but which trade unions would like to bargain on...................................................... 13 Figure 9: AI-related topics that trade unions do not wish to bargain on......................................................................... 14 Figure 10: AI-related topics included in CBAs................................................................................................................... 15 Figure 11: Except collective bargaining, what tools and measures does your trade union use to raise awareness and train members................................................................................................................................................. 16 Figure 12: Number of CBA referring to the introduction of AI and/or AM at work........................................................... 18 Figure 13: Sectoral distribution of the collective agreements referring to the introduction of AI and/or AM at work........ 18 Figure 14: Percentage of agreements analysed referring to specific topics........................................................................ 19 Figure 15: Artificial Intelligence(AI) and/or Algorithmic Management(AM) at work included in collective agreements in the different countries................................................................................................................................. 20 Figure 16: Position of the respondent in the trade union.................................................................................................. 26 Figure 17: Involvement of the trade union of the respondent in the collective agreement................................................ 26 Figure 18: Trade unions and their bargaining counterparts by country............................................................................. 27 Figure 19: Level at which the collective bargaining agreement was concluded?............................................................... 28 LIST OF TABLES Table 1: Reasons for the absence of AI-related provisions in CBAs................................................................................... 16 Table 2: Relevant variables included in the WageIndicator Collective Agreements Database annotation system............... 17 Table 3: Number of agreements that include clauses on the relevant topics and the years when the agreements commenced........................................................................................................................................................ 19 Table 4: Number of responses by country......................................................................................................................... 25 Table 5: The number of responses in the survey by sector................................................................................................ 26 Table 6: Trade unions and their bargaining counterparts by country................................................................................. 27 Table 7: Examples of other CBAs that contain AI-related provisions(as reported by survey respondents)......................... 28 LIST OF ABBREVIATIONS AAMS AI AM CB CBA GDPR SD TU Artificial intelligence and Algorithmic management systems Artificial intelligence Algorithmic management Collective bargaining Collective bargaining agreement General Data Protection Regulation Social dialogue Trade union 2

Executive Summary Executive Summary Artificial intelligence(AI) systems are increasingly being used in various industries in connection with algorithmic man­ agement, chatbots, geopositioning, and other processes. AI refers to machine-based systems that can make predictions, recommendations, or decisions with only limited human input/­oversight. To understand the challenges that are emerging in relation to the increased use of AI in human resource management, the report examines the current situation in collective bargain­ ing regarding the use of AI-related tools by employers vis-àvis workers, especially in the service sector. The findings are based on desk research, an original survey of 148 trade union representatives affiliated to UNI Europa in 32 countries, and an analysis of 31 collective agreements that already contain provisions relating to the use of AI. Results reflect current experience, general opinions on bargaining on AI-related challenges, and expected union actions to develop bargain­ ing in this area, as well as some good practices on AI-related clauses in collective agreements. The analysis has produced the following key conclusions: – Bargaining on AI is emerging and is not yet as widespread as bargaining on other elements of working conditions. – Out of 90 survey responses, only 20% of trade unions reported having a collective agreement that addresses AI-related issues at the organisation or sector level. This implies that the majority of trade unions(69%) do not have any collective bargaining agreements related to AI, and 11% are unaware of any such agreements. – Existing collective agreements mostly make general ref­ erence to the use of technology. However, several agree­ ments were identified(e.g. in Italy, Germany, Norway and Spain) that can serve as examples of more detailed rules and arrangements on the right to disconnect, digital rights of the workers at the workplace, informa­ tion-sharing and business control. – With the increasing use of technology at the workplace, it can be expected that collective bargaining on AI will further increase in relevance. 42% of the UNI Europa affiliates that participated in the survey are already en­ gaged in discussions and negotiations on various topics related to AI, even if this is not collective bargaining in the strict sense. Unions prioritise issues of data protec­ tion, worker privacy, the impact of AI on working hours, monitoring of worker activities, and automated sched­ uling of work shifts. – In the process of developing collective bargaining on AI-related issues, unions prioritise bargaining on work­ ers’ right to challenge decisions made through auto­ mated decision-making, and their right to receive advice from an external data expert. Additionally, there is an in­ tense desire among unions to have a right to information and consultation on the use and evaluation of AI tools. 3

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Introduction Artificial intelligence(AI) systems are increasingly being used across various industries for algorithmic management, chat­ bots, geopositioning, and other processes. The definitions of AI vary. One definition is that AI is any machine-based system that can make predictions, recommendations, or decisions with only limited human input / oversight. The High-Level Expert Group on Artificial Intelligence has developed another much more complex definition: To address the above challenges, UNI Europa and the FES Competence Centre on the Future Work are actively rais­ ing awareness among workers and unions. Earlier work (Holubová et al. 2022) highlighted the need for gathering, analysing and presenting specific examples of how collective agreements as an important governing tool to protect work­ ers are able to address AI-related challenges. In responding to the above needs, the current report has a twofold aim: »Artificial intelligence systems are software(and possibly also hardware) systems designed by humans that, given a com­ plex goal, act in the physical or digital dimension by perceiv­ ing their environment through data acquisition, interpreting the collected structured or unstructured data, reasoning on the knowledge, or processing the information, derived from this data and deciding the best action(s) to take to achieve the given goal. AI systems can either use symbolic rules or learn a numeric model, and they can also adapt their be­ haviour by analysing how the environment is affected by their previous actions“(High-Level Expert Group on Artificial Intelligence, 2019). In some cases, AI is interchangeably referred to as algorithmic management. There is a certain scepticism regarding the true nature of some of the technology tools that are labelled as AI. There is some doubt regarding whether they are actually AI or merely simple algorithmic management tools. In any case, AI algorithmic management systems(AAMS) in the workplace give rise to concerns about data protection, pri­ vacy, power relations, and human rights. In particular, AAMS threaten to“illegitimately surveil workers and their personal data, create a divide in access to knowledge between man­ agers and workers, to speed up work to(or past) breaking point, and to take operational decisions including perfor­ mance management, worker remuneration and even hiring and firing without sufficient human oversight. Furthermore, AI and algorithmic management tools are often ›black-box‹ technologies which use machine-learning techniques to eval­ uate data and make recommendations and decisions. The study is part of a broader research agenda that is being pursued by Friedrich Ebert Stiftung(FES) and UNI Europa exploring the growing significance of AAMS for European workers as well as prospects for collective bargaining to as­ sert worker control over these new and proliferating tools. – Presenting the current situation in collective bargaining regarding the use of AI-related tools by employers vis-àvis workers, especially in the services sector. The research team carried out a survey among UNI Europa members and presents here the results in terms of current expe­ rience, general opinions on bargaining on AI-related challenges, and expected union actions to develop bar­ gaining in this area. – Analysis of 31 existing collective agreements that already contain provisions relating to the use of AI and present­ ing these as good practice for further use and dissemina­ tion among trade unions. The clauses analysed address topics such as data protection, surveillance, recruitment or automatic work organisation, as well as rules to gov­ ern the introduction of new technologies. To perform this task, the report uses the following method­ ology: – Desktop research to identify existing publications and literature on the relevance of AI in collective bargaining; – A survey among UNI Europa sector affiliates to gather data on collective agreements they have negotiated on AI-related rules and arrangements; – Analysis of the content of existing collective agreements that contain AI-related provisions. The research results have expanded an already existing global collective agreement database, while the collective agreements collected have been coded on various AI-related variables. This allows national trade unions to access best practices across European service sectors and use these stip­ ulations and provisions in their own negotiations. For the purpose of the project, especially the survey, the definition of AI was simplified and related to a diverse set of technological tools and techniques to remotely manage 4

workforces, relying on data collection and surveillance of workers to enable automated or semi-automated decision-­ making(Mateescu& Nguyen, 2019). The report is structured as follows. First, it provides a brief overview of the literature on the use of artificial intelligence (AI) in collective bargaining(CB), collective bargaining agree­ ments(CBA), or any other related tools. The second chapter presents the results of an online survey conducted among UNI Europa affiliates from April till September 2023. The sur­ vey aimed to assess the relevance of AI-related bargaining and to identify CBAs with AI-related stipulations. Finally, the third chapter summarises the findings from coding and ana­ lysing the content of 31 CBAs that contain AI-related clauses. The report concludes with an overview of the findings and lessons learned. Introduction 5

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors 1 CURRENT KNOWLEDGE ON AI IN COLLECTIVE BARGAINING AND COLLECTIVE AGREEMENTS As the landscape of modern industries evolves, an increasing number of workplaces are embracing the capabilities and potential of artificial intelligence(Schwartz et al. 2023). Al­ gorithmic management systems are employed for a wide range of tasks, from recruitment and training to scheduling and performance assessments. Moreover, they are integral in automating tasks previously handled by human workers, despite the fact that these algorithms may lack sophistica­ tion. These systems are often marketed as tools that enhance efficiency and objectivity by reducing the impact of human subjectivity(Schwartz et al. 2023). However, it is not uncom­ mon for these algorithms to perpetuate and even magnify existing disparities and biases. De Stefano and Taes(2023) argue that concerns include the potential for algorithmic bias, unjust handling of employee data, and encroachment upon personal privacy. Monitoring tools might encroach on workers’ activities beyond their designated work hours, and the constant surveillance could amplify psychological and social risks. The integral aspect of safeguarding workers’ rights and achieving a more equitable distribution of advantages stem­ ming from novel AI-driven management tools lies in collec­ tive bargaining and proactive initiatives undertaken by trade unions(De Stefano& Taes, 2023; Aloisi& De Stefano, 2021). Given the technology’s early stage of development, limited practical experience, and significant uncertainties surround­ ing its future evolution, trade unions are currently tackling the urgent and challenging task of elaborating new legisla­ tion and ethical guidelines for AI utilisation at the workplace (Krzywdzinski et al. 2022). Trade union representatives do not have to acquire extensive expertise in all emerging tech­ nologies to effectively represent their members. Instead, their primary objective should revolve around recognising possible problems or worries for both current and prospective mem­ bers. They should then advocate for employers to provide the necessary assurances, information, explanations, and proto­ cols that instil confidence among employees regarding the proper and responsible use of these technologies(Prospect, 2021). 1.1 EXAMPLES FROM ITALY AND THE UK AI is already being included in collective bargaining discus­ sions in several countries. Numerous initiatives have been undertaken to facilitate the process of collective bargaining and the subsequent conclusion of collective agreements. In 2018, the trade unions FILCAMS-CGIL and FISASCAT-CISL in Italy reached an agreement with Partesa, a subsidiary of the Heineken Group, regarding the utilisation of the Tele­ matics smartphone application. This app is designed to over­ see drivers’ adherence to regulations, enhance safety, and respond to drivers’ conduct and emotional states. According to legal provisions, the use of monitoring tools is restricted to specific purposes, including achieving productivity targets, ensuring safety, and safeguarding company assets. Further­ more, such equipment must receive prior approval from the appropriate trade union or works council before being imple­ mented(UNI Global Union, 2023). Another good example of reaching an agreement regarding algorithmic management is the case of the GMB union and the company Hermes in the United Kingdom. Negotiated in 2019, this agreement per­ tains to some of the algorithm-based procedures employed by the company to oversee their fleet of delivery drivers. While it does not constitute a comprehensive negotiation on the company’s algorithmic management practices, it requires the company to modify its automated payment system to guarantee that workers receive at least the minimum wage and promptly receive any bonuses earned without needing to retroactively request them. Additionally, the agreement empowers unions to conduct health and safety assessments in the event of incidents, enabling them to highlight instances where algorithmic management practices pose safety con­ cerns. It also introduces a mechanism for workers to contest decisions made by technology(Collins& Atkinson, 2023). After almost a year of disputes, the Communication Workers Union and the Royal Mail Group in the United Kingdom also successfully struck a deal on algorithmic management. In the segment labelled ›Technology,‹ provisions have been established to protect current wage and work hours in the event of new procedures being implemented, ensuring that human decision-making remains a central component of the company’s operations(RMG& CWU, 2020). In their position paper regarding the use of AI systems, the Nordic Financial Unions also highlighted that it is imperative that they be purposefully designed and deployed to function as comple­ mentary tools aimed at enhancing human capabilities, rather than as substitutes intended to replace them(Nordic Finan­ cial Unions, 2022). 6

Current knowledge on AI in collective bargaining and collective agreements 1.2 EXAMPLE FROM SPAIN In Spain, the leading telecommunications company, Tele­ fonica, signed a national agreement on the right to discon­ nect with the trade unions representing its employees, UGT and Comisiones Obreras(Telefónica, 2018). There have also been efforts throughout the EU to shape public policy to­ ward AI in regulating work. Spain can be considered a lead­ ing country in the European Union in the volume of work conducted via platforms. Spain implemented a pioneering law on algorithmic management in 2021. Law 12/2021, better known as Ley rider(the Riders’ Law), is based on an agreement between the trade union CCOO(Workers’ Commission) and UGT(General Workers’ Confederation), employer organisation CEOE(Spanish Confederation of Busi­ ness Organisations) and CEPYME(Spanish Confederation of Small and Medium Enterprises), and the Spanish government (Rodríguez, 2022; Eurofound, 2021). The recently enacted legislation essentially fills in the gaps of the GDPR Law, which left out collective and trade union aspects of the workplace. Empowering trade unions, the law states that companies must inform workers’ representatives about the criteria, reg­ ulations, and instructions governing algorithms or artificial intelligence systems that impact decision-making, potentially affecting working conditions and access to employment, in­ cluding profiling. This implies that the works council has the right not only to know if there is an algorithm influencing working conditions and employment, but also to understand the specifics of how the algorithm operates(Todolí-Signes, 2021). 7

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors 2 FINDINGS OF THE SURVEY AMONG UNI EUROPA AFFILIATES The online ›Survey on Collective Bargaining Agreements Re­ lated to Artificial Intelligence‹ is aimed at trade union officials or delegates directly involved in collective bargaining or with in-depth knowledge about collective agreements among UNI Europa affiliates. The survey looked at the relevance of AI in collective bargaining, existing CBAs with clauses related to AI, and reasons for the absence of AI clauses in CBAs. The survey also aimed to discover other tools and measures re­ lated to AI aside from CBAs. The survey was not anonymised to identify the trade unions and their opinion and experience with AI-related CB. The data were collected over the period 24 May 2023–19 September 2023. The survey was provided in 6 languages (EN, RO, IT, DE, ES, FR) to increase the response rate. The respondents were allowed to choose the language. The UNI Europa representatives sent the link to the survey to affiliates in 43 countries in the EU and beyond with repeated remind­ ers along with a request to participate in the survey. The survey targeted mostly the service sector, particularly agency work, care, cleaning and security, commerce, finance, ICT services, media and entertainment and gaming, post and logistics, and tourism. The total number of responses was 148 from trade union­ ists from 32 countries. The countries most represented in the analysis are France(14%), Belgium(13%), Denmark (6%), Italy(6%) and Norway(6%). Other countries’ re­ sponses comprised less than 6%. 1 As to the position of the trade unionists involved in the sur­ vey, half of the respondents were national trade union officers(49%). Sectoral trade union officers comprised 28% and local / regional 13%. Company shop stewards or delegates were least represented in the survey, at 11%. Regarding the respondents’ involvement in collective bar­ gaining, most directly negotiated collective agreements (71%), while 22% were not directly involved in collective bargaining, but were informed about bargaining activities. The rest were members of broader bargaining teams or ad­ visors. Respondents not involved in bargaining accounted for 3% of the 148 total respondents. After re-categorisation of the open questions, the sectors most represented in the sample were the ICTS sector, finance, post and logistics, commerce and media. The option ›other‹ comprised sectors not covered by UNI Europa, but by other trade union federations(i.e. metallurgy and public sector). 2.1 THE OVERALL RELEVANCE OF AI-RELATED COLLECTIVE BARGAINING The survey explored how relevant AI topics are for CB. For 62% of the respondents, using AI systems is a relevant topic in their CA. 47% of the trade unions in the sample addressed AI-related topics in collective bargaining compared to 22% when the employer took the initiative with this topic. Another 47% of the respondents state that using AI systems is a relevant topic in collective bar­ gaining for the employers with whom they bargain. 42% of the respondents confessed that they are conducting bargaining on various topics related to AI.[Figure 1] 2.2 TOPICS COVERED IN AI-RELATED COLLECTIVE BARGAINING The survey results focus on the topics of AI-related bargain­ ing, such as working time, training on AI tools, workforce and job quality, workers’ rights, and recruitment and or­ ganisation of work. The survey also examines the perceived impact and opinions on these topics based on whether the trade unions already bargain on AI-related topics or wish to do so and those who do not bargain and have no interest in doing so. Those who already bargain on AI view the largest impact of AI to be on working time(e.g. the right to discon­ nect)(27%), training for the staff on new AI tools(22%) and workload and work intensity(22%). Out of those who do not bargain but wish to do so, the greatest impact is perceived to be on staffs’ training in new AI tools(60%), training for management(59%), and workload and work intensity (57%). Trade unions which do not bargain on AI and do not wish to do so perceive most often the impact of AI to be on the number of jobs and social plans(19%) and the smallest impact on training staff to use new AI tools(9%).[Figure 2] 1 For a detailed description of the sample, see the Annex 8

Findings of the survey among UNI Europa affiliates Figure 1: Relevance of the AI-related collective bargaining(N=90, in%) For our union, the use of AI systems is a relevant topic in collective bargaining. For the employer(s) with whom we bargain, the use of AI systems is a relevanttopic in collective bargaining. Our trade union initiated AI-related topics in collective bargaining 62 47 47 30 39 46 Our trade union conducts bargaining on themes related to AI. The employer(s) who is/are our bargaining partner(s) initiated AI-related topics incollective bargaining. 42 46 22 70 0 10 20 30 40 50 60 70 80 90 100 Agree Disagree Don’t know Q7: Express your opinion on the following statements on the scale: strongly agree, agree, disagree, strongly disagree and don´t know. Note: Strongly agree and agree were merged into agree; disagree and strongly disagree into disagree Figure 2: Perceived impact of AI-related bargaining on the workforce, job quality and training(N=90, in%) Impact of AI systems on working time/ right to disconnect 27 54 10 Training for staff on new AI tools 22 60 9 Impact of AI systems of workload/work intensity 22 57 11 Training for management on new AI tools Impact of deployment of AI tools on number of jobs/ headcounts; necessity to negotiate social plans 18 59 14 18 54 19 0 10 20 30 40 50 60 70 80 90 100 We bargain on this topic We do not bargain on this topic but we wish to do so We do not bargain on this topic and we do not wish to do so Don’t know Q8: Please share your opinion on the impact of AI on the workforce regarding the number and quality of jobs and training needs in your collective bargaining. Specify what best applies for each of the following statements below. Respondents could also cite other kinds of impact of bargain­ ing on AI-related topics: – AI is primarily negotiated in company agreements with works councils at the company level. The AI topic is increasingly coming into focus. – General discussions on when AI is a good tool and when not and how to implement and interact with it. – Impact on transparency. – Data Ethics with regards to AI. – AI does not replace personnel in our industry. – If we cannot block the use of AI tools, we ask to be able to negotiate its mode of use. – The company is now preparing work with AI tools so that it may become relevant in future, but it has not been a topic yet. – Employers who use AI only for profit without thinking about employees’ health(mental and physical). Regarding the impact of AI-related bargaining on union rights, responses differ between those who already bargain and those who wish to bargain. Those who bargain on AI see the strongest impact on the TU in­ volvement in data protection and impact assessment(33%) and the right to information on AI tools(31%). Conversely, those TU who only wish to bargain on AI see the highest impact on the right to repeal a decision based on automated decision-making(68%) and be advised by external data ex­ perts(60%).[Figure 3] The open questions teased additional information from the respondents: – At the company level, more and more rules and arrangements / company agreements are being concluded. – Negotiations here are not based on a collective agreement, but rather a company agreement. 9

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 3: Perceived impact of AI-related bargaining on trade unions’ rights(N=90, in%) Worker/trade union involvement in data protection impact assessment 33 48 11 Right to information/consultation on the use and evaluation of AI tools Right to information/consultation on design of AI tools(before introduction) Right to be advised by external data expert 31 23 19 52 9 59 9 60 12 Right to repeal a decision based on automated decisionmaking 13 68 9 0 10 20 30 40 50 60 70 80 90 100 We bargain on this topic We do not bargain on this topic but we wish to do so We do not bargain on this topic and we do not wish to do so Don’t know Q9: Please share your opinion on trade union rights related to AI in your collective bargaining. Specify what best applies for each of the state­ ments below. Figure 4: Opinion on the use of AI-related bargaining on the recruitment process, work organisation and workers’ assessments (N=90, in%) Scheduling work shifts Automated worker’s assessment against output and performance targets Automated worker assessment based on consumer ratings Team/task allocation, algorithm to distribute orders Algorithmic management systems to approve/deny annual leaves, log sick leave etc. Automated background check of applicants on social media Automated CV screening or job interviews 23 49 18 22 51 16 16 51 20 14 54 20 13 56 22 8 59 21 8 58 20 0 10 20 30 40 50 60 70 80 90 100 We bargain on this topic We do not bargain on this topic but we wish to do so We do not bargain on this topic and we do not wish to do so Don’t know Q10: Please share your opinion on using AI in the recruitment process, work organisation and worker´s assessment in your collective bargain­ ing. Specify what best applies for each of the following statements. – If we cannot block the use of AI tools, we ask to be able to negotiate its mode of use. – We have strong national privacy regulations. In the next stage, the survey asked about the use of AI in the recruitment process, work organisation and worker assess­ ment in collective bargaining of the trade unions. 23% of re­ spondents stated that they bargain on using AI in scheduling work shifts and automated workers’ assessments in terms of output and performance targets(22%). Those who do not bargain, but wish to do so, consider the most relevant bar­ gaining on the automated background check of applicants on social media(59%) and automated CV screening or job interviews within the recruitment process(58%).[Figure 4] Additional information provided by the respondents: – As of today, these are irrelevant, as the companies have not wished to implement any of these, but we are prepared for this to be an issue in the future and will then demand to be involved in the discussion before as well as conditions underlying implementation. I have therefore selected the option»We bargain on this topic.« – We work with people on people. – The works council is attending to these topics. 10

Findings of the survey among UNI Europa affiliates Figure 5: Opinion on the use of AI tools for worker surveillance in CB(N=90, in%) Software to track physical or digital worker activity(e.g.… 26 46 20 Access to employee emails and messages 22 42 27 Wearables/devices using sensors to collect information 18 54 20 Voice analytics(e.g. in contact centres) 14 47 23 Systems to generate automatic responses(e.g. chatbots) 11 50 24 AI systems for language or visual recognition& processing 11 52 21 0 10 20 30 40 50 60 70 80 90 100 We bargain on this topic We do not bargain on this topic but we wish to do so We do not bargain on this topic and we do not wish to do so Don’t know Q11: Please share your opinion on AI tools used for worker surveillance and in other cases in your collective bargaining. Specify what best ap­ plies for each of the following statements. Figure 6: Opinion on data protection in CB(N=90, in%) Respecting the relevant privacy/data protection legislation Collecting, processing, storing and using workers’ personal data Restrictions on sharing workers’ personal data Provision of information about the AI systems in plain, understandable language 0 43 41 39 20 10 20 30 34 38 40 58 40 50 60 70 16 14 13 13 80 90 100 We bargain on this topic We do not bargain on this topic but we wish to do so We do not bargain on this topic and we do not wish to do so Don’t know Q12: Please share your opinion on the topic of data protection in your collective bargaining. Specify what best applies for each of the follow­ ing statements. Opinions on the use of AI tools for workers’ surveillance in CB revealed that trade unions already bargain mostly on soft­ ware to track physical or digital workers’ activity(26%) and access to employee email and messages(22%). Conversely, those who wish to bargain on AI perceive as most relevant bargaining on wearables / devices using sensors to collect information(54%), and among those who already bargain on AI, on AI systems for language or visual recognition and processing(52%).[Figure 5] topic of provision of information about AI systems in plain, understandable language(58%).[Figure 6] Additional information: – Laws and rules related to GDPR and privacy give us options. – Not an issue in CB. – The works council attends to these topics, and we have sufficient privacy rules and arrangements. The last cluster of work-related aspects that are connected to data protection produced the following results. Data protec­ tion is the most prevalent topic that the trade unions in the survey are already bargaining on. Most focus on respecting relevant privacy/data protection legislation(43%) and collec­ tion, processing, storing and using workers’ personal data. Those who wish to bargain on AI attach relevance to the 2.3 RELEVANCE OF PARTICULAR TOPICS IN AI-RELATED COLLECTIVE BARGAINING Merging all AI-related topics in one list allows us to inspect the relevance of AI-related topics separately for those trade unions which already bargain on the topic, those which do 11

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 7: Bargained AI-related topics by trade unions(N=507, in%) We bargain on this topic Respecting the relevant privacy/data protection legislation Collecting, processing, storing and using workers’ personal data Restrictions on sharing workers’ personal data Right to be advised by external data expert Impact of AI systems on working time/right to disconnect Software to track physical or digital worker activity(e.g. geolocation, key stroke monitoring) Scheduling work shifts Access to employee emails and messages Automated worker’s assessment against output and performance targets Right to information/consultation on the use and evaluation of AI tools Right to information/consultation on design of AI tools(before introduction) Training for staff on new AI tools Impact of AI systems of workload/work intensity Provision of information about the AI systems in plain, understandable language Wearables/devices using sensors to collect information Right to repeal a decision based on automated decision- making Worker/trade union involvement in data protection impact assessment Training for management on new AI tools Impact of deployment of AI tools on number of jobs/headcounts; necessity to negotiate social plans Automated worker assessment based on consumer ratings Voice analytics(e.g. in contact centres) Team/task allocation, algorithm to distribute orders Algorithmic management systems to approve/deny annual leaves, log sick leave etc. Systems to generate automatic responses(e.g. chatbots) AI systems for language or visual recognition& processing Automated background check of applicants on social media Automated CV screening or job interviews 4,7 4,7 4,5 4,1 3,9 3,9 3,9 3,9 3,9 3,9 3,6 3,2 3,2 3,2 3,2 3,2 2,8 2,6 2,6 2,4 2,0 2,0 1,4 1,4 7,7 7,3 6,9 Note: Merged responses for Q9–Q12 for those unions which bargain on the AI-related topic not do so, but which wish to bargain on the specific topic, and those which do not wish to bargain on the topic. AI-related topics that are currently receiving the most attention from trade unions are centred on data protection and privacy for workers. The next most com­ monly bargained topics are the effects of AI on working hours, monitoring of worker activity, and automated sched­ uling of work shifts. The least bargained topics among un­ ionists are related to the hiring process.[Figure 7] The order of AI-related topics that have not been bargained yet, but are desired as bargaining topics, differ from that that have already been bargained. The order of the topics 12

Findings of the survey among UNI Europa affiliates Figure 8: AI topics not bargained, but which trade unions would like to bargain on(N=1286, as a%) We do not bargain on this topic but we wish to do so Right to repeal a decision based on automated decision- making Right to be advised by external data expert Training for staff on new AI tools Automated background check of applicants on social media Right to information/consultation on design of AI tools(before introduction) Training for management on new AI tools Automated CV screening or job interviews Impact of AI systems of workload/work intensity Algorithmic management systems to approve/deny annual leaves, log sick leave etc. Respecting the relevant privacy/data protection legislation Wearables/devices using sensors to collect information Team/task allocation, algorithm to distribute orders Impact of AI systems on working time/right to disconnect Impact of deployment of AI tools on number of jobs/headcounts; necessity to negotiate social plans AI systems for language or visual recognition& processing Right to information/consultation on the use and evaluation of AI tools Automated worker’s assessment against output and performance targets Automated worker assessment based on consumer ratings Systems to generate automatic responses(e.g. chatbots) Scheduling work shifts Worker/trade union involvement in data protection impact assessment Provision of information about the AI systems in plain, understandable language Voice analytics(e.g. in contact centres) Collecting, processing, storing and using workers’ personal data Software to track physical or digital worker activity(e.g. geolocation, key stroke monitoring) Restrictions on sharing workers’ personal data Access to employee emails and messages 4,7 4,2 4,2 4,1 4,1 4,1 4,0 4,0 3,9 3,8 3,8 3,8 3,8 3,8 3,7 3,7 3,6 3,6 3,5 3,4 3,3 3,3 3,3 3,2 3,2 3,0 3,0 Note: Merged responses for Q9–Q12 for those who do not bargain, but would like to do so, on the AI-related topic can also be perceived as a ›wish list‹ that the trade unions would like to bargain in the future and regarding which they would like to have concrete provisions in collective bargain­ ing agreements. The most requested topic is related to the rights of workers, specifically the right to challenge decisions made through automated decision-making and the right to receive advice from an external data expert. Additionally, there is a strong desire for a right to information and consultation about the use and assessment of AI tools. Other important topics include staff training on new AI tools and automated background checks of social media profiles for job applicants.[Figure 8] 13

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 9: AI-related topics that trade unions do not wish to bargain on(N=398, as a%) We do not bargain on this topic and we do not wish to do so Access to employee emails and messages Systems to generate automatic responses(e.g. chatbots) Voice analytics(e.g. in contact centres) Algorithmic management systems to approve/deny annual leaves, log sick leave etc. AI systems for language or visual recognition& processing Automated background check of applicants on social media Wearables/devices using sensors to collect information Software to track physical or digital worker activity(e.g. geolocation, key stroke monitoring) Automated worker assessment based on consumer ratings Team/task allocation, algorithm to distribute orders Automated CV screening or job interviews Impact of deployment of AI tools on number of jobs/headcounts; necessity to negotiate social plans Scheduling work shifts Respecting the relevant privacy/data protection legislation Automated worker’s assessment against output and performance targets Collecting, processing, storing and using workers’ personal data Training for management on new AI tools Provision of information about the AI systems in plain, understandable language Restrictions on sharing workers’ personal data Right to be advised by external data expert Worker/trade union involvement in data protection impact assessment Impact of AI systems of workload/work intensity Impact of AI systems on working time/ right to disconnect Right to repeal a decision based on automated decision- making Right to information/consultation on the use and evaluation of AI tools Right to information/consultation on design of AI tools(before introduction) Training for staff on new AI tools 6,0 5,5 5,3 5,0 4,8 4,8 4,5 4,5 4,5 4,5 4,5 4,3 4,0 3,5 3,5 3,3 3,3 3,0 3,0 2,8 2,5 2,5 2,3 2,0 2,0 2,0 2,0 Note: Merged responses for Q9–Q12 for those who do not bargain and do not wish to develop bargaining on AI-related topics The topics that are neither being bargained nor for which there is a desire to bargain on them may indicate the sec­ ondary relevance of AI-related issues for trade unions. On the other hand, it may indicate that AI-related topics are not seen as a threat to workers so far, or that trade unions have limited information on these matters. The least relevant topics in AI-related bargaining include ac­ cess to employee emails and messages, systems for generat­ ing automatic responses(e.g. chatbots), and voice analytics (e.g. in contact centres).[Figure 9] 14

Findings of the survey among UNI Europa affiliates 2.4 EXISTING COLLECTIVE BARGAINING AGREEMENTS WITH AI-RELATED TOPICS Out of the 90 responses, 20% of trade unions stated that there is a collective agreement related to AI at the level of their organisation or sector. This is only a minority of the trade unions, as 69% of those which responded in the survey do not have any CBA, and 11% do not know. The level of the CBA, the sectoral and multi-employer level, and even organisation level evenly account for 25%, respec­ tively. A CBA at one workplace was only cited in 6% of responses. Figure 10: AI-related topics included in CBAs(N=176, as a%) Respecting the relevant privacy/data protection… Collecting, processing, storing and using workers’… Restrictions on sharing workers’ personal data Impact of AI systems on working time/right to disconnect Worker/trade union involvement in data protection… Training for staff on new AI tools Right to information/consultation on the use and… Right to repeal a decision based on automated decision-… Training for management on new AI tools Voice analytics(e.g. in contact centres) Access to employee emails and messages Automated worker’s assessment against output and… Team/task allocation, algorithm to distribute orders Scheduling work shifts Right to information/consultation on design of AI tools… Right to be advised by external data expert Impact of AI systems of workload/work intensity Systems to generate automatic responses(e.g. chatbots) AI systems for language or visual recognition& processing Automated worker assessment based on consumer… Algorithmic management systems to approve/deny… Impact of deployment of AI tools on number of… Provision of information about the AI systems in plain,… Wearables/devices using sensors to collect information Software to track physical or digital worker activity(e.g.… Automated background check of applicants on social… Automated CV screening or job interviews 7,4 7,4 6,8 6,3 5,7 5,7 4,5 4,0 4,0 3,4 3,4 3,4 3,4 3,4 3,4 3,4 3,4 2,8 2,8 2,8 2,8 2,8 2,3 1,1 1,1 1,1 1,1 Q15–Q19: Which of the following topics are mentioned in the collective bargaining agreement? 15

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 11: Except collective bargaining, what tools and measures does your trade union use to raise awareness and train members (N=86, as a%) Consultation to facilitate new legislation(e.g. the right to disconnect) 49 Policy/strategy papers on AI 36 Trainings/workshops addressing the risks of AI 36 Social partner initiatives(e.g. joint statements, projects) 30 Guidelines on data impact assessment 30 Other(please specify) 22 Checklists on ethical AI 21 Checklists or Guidelines on the introduction and use of AI at work 19 Q25: Except collective bargaining, what tools and measures does your trade union use to raise awareness and train members in AI, and to minimise the risks related to AI usage? Respondents also added additional comments: – National basic agreement between employers and union federations – At national sectoral, company level and sometimes at the workplace level – We have not signed anything with the employer yet The topics mentioned in the CBA reflect AI-related topics that have also been bargained by the unionists. The most frequent provisions related to AI in CBAs are those connected to the protection of personal data and privacy of workers: Restrictions on sharing workers’ personal data, collecting, processing, storing and using workers’ per­ sonal data and respecting relevant privacy/data protection legislation.[Figure 10] Table 1: Reasons for the absence of AI-related provisions in CBAs The reasons AI has only just begun to be discussed or bargained, and to compile a CBA Not necessary yet, there are other pri­ orities Employer refused to discuss AI Lack of information on AI so far Lack of power and resources or respec­ tive legislation Using broader regulation to mitigate the threat of AI Lack of transparency of AI usage in the company Number of related responses 17 12 7 5 5 4 3 Several respondents provided the full text of specific provisions. These are analysed in Section 3.2 of this report, which is specially devoted to these. Additional responses were ›do not know‹ or complete ab­ sence of collective bargaining. 2.5 REASONS FOR THE ABSENCE OF AI-RELATED PROVISIONS IN COLLECTIVE AGREEMENTS The survey explored the reasons for the absence of provisions relating to AI in collective bargaining agreement(s)(Q24). Out of more than 60 open answers, the most prevalent rea­ sons can be categorised as follows: 2.6 OTHER TOOLS AND MEASURES RELATED TO AI Except for CB, trade unions mostly use consultation to facili­ tate new legislation(49%), policy strategy papers on AI(36) and training / workshops addressing the risks of AI(36%) (see Figure 11). Checklists or guidelines on introducing and using AI at work are used least according to the current sur­ vey(19%). 16

Annotation, coding and analysis of collective agreements’ contents 3 ANNOTATION, CODING AND ANALYSIS OF COLLECTIVE AGREEMENTS’ CONTENTS In addition to the survey analysis, the authors collected 31 CBAs that contain AI-related provisions. These CBAs were included in the already existing global WageIndicator CBA database. A dedicated system of codes has been developed (see Table 2); and these codes were applied to the analysis of the 31 CBAs that were studied. CBAs considered for the analysis among the different coun­ tries. The country with the most CBAs in the sample is Spain (15 CBAs in the sample were collected there). The sample includes 8 CBAs from Germany, 3 from Italy, while there are four countries with one agreement each(Finland, Norway, Switzer­land and the UK). The following map and graphs take into account agree­ ments that specifically refer to the introduction of Artificial Intelligence(AI) and/or Algorithmic Management(AM) at work( newtech_aiam= 1). As one agreement has been ex­ cluded, the number of agreements in the final sample is 30. The map in Figure 12 shows the geographic distribution of Looking at the sectoral distribution of CBAs in our sample, manufacturing(9 CBAs) and ICT services(6 CBAs) are the sectors with the most CBAs, followed by financial services, banking and insurance(4 CBAs). Two of the agreements(one from Norway and one from Spain) do not have a sector of reference(see Figure 13). Table 2: Relevant variables included in the WageIndicator Collective Agreements Database annotation system Question ID newtech_trigger newtech_aiam newtech_topics Question Label Does the agreement make any reference to the intro­ duction of advanced technologies(including Artificial Intelligence and Algorithmic Management) at work? * By AI(Artificial Intelligence) we refer broadly to any machine-based system that can make predictions, rec­ ommendations, or decisions with only limited human input / oversight. By AM(Algorithmic Management) we mean the del­ egation of managerial functions to algorithmic and automated systems. Does the agreement specifically refer to the introduc­ tion of Artificial Intelligence(AI) and/or Algorithmic Management(AM) at work? In which of the following topics is AI/AM mentioned? You can select more than one option. Please be as specific as possible. Answers(code and label) 0= No 1= Yes 0= No 1= Yes 11= Impact of AI/AM systems on working time and the right to disconnect 22= Training for employees and/or management on new AI tools(including the risks related to AI usage) 33= Use of AI/AM in the recruitment process, work organi­ sation and assessments of workers 44= Employee/trade union involvement in data protection 66= Employee/trade union involvement when new tech­ nologies are introduced 77= Respecting the relevant privacy/data protection leg­ islation 88= AI/AM tools used for monitoring and worker sur­ veillance(e.g., software/devices to track physical or digital worker activity) 55= Other topics –9= Not specified 17

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 12: Number of CBA referring to the introduction of AI and/or AM at work Number of CBAs referring to the introduction of Al and/or AM at work 0 15 Norway 1 Sweden Finland 1 United Kingdom 1 Germany 8 Spain 15 Switzerland France 1 Italy 3 Poland Belarus Ukraine Source: WageIndicator Collective Agreements Database, accessed in November 2023. N=30 Figure 13: Sectoral distribution of the collective agreements referring to the introduction of AI and/or AM at work. Sector Manufacturing 9 ICT services 6 Finan cial services, banking, insurance 4 Transport, logistics, communication 3 Not specified 2 Hospitality, catering, tourism 2 Retail trade 1 Publishing, printing, media 1 Legal and market consultancy, business activities 1 Agriculture, forestry, fishing 1 Source: WageIndicator Collective Agreements Database, accessed in November 2023. N=30 3.1 ANALYSIS OF CBAS Figure 14 shows the percentage of agreements that include clauses on the specific topics. The most common AI-related topic addressed in the CBAs analysed include training for employees and/or management on new AI tools (22 agreements) and employee / trade union involvement when new technologies are included(18 agreements). The analysis of the type of AI-related provisions and the year in which the relevant CBAs entered into force(see Table 3) show that it was indeed the latest years, after 2021, when AI-related topics became increasingly relevant. This holds especially true when one zooms in on provisions govern­ ing training for employees and/or management on new AI tools. The involvement of employees / trade unions when new technologies are launched was addressed three times more often in 2022 than in the year before and the year after. The impact of AI/AM systems on working time and the right to disconnect have also gained more attention in the last two years, mostly in 2023. This reflects how EU-level policymaking has spurred collective bargaining in particular Member States and sectors, since the right to disconnect was extensively addressed in EU-level policy discourses. Other is­ sues, e.g. data protection legislation, or the use of AI for monitoring surveillance of workers, have not undergone in­ creased attention in the past years. 18

Annotation, coding and analysis of collective agreements’ contents Figure 14: Percentage of agreements analysed referring to specific topics. Training for employees and/or management on new Al tools (including the risks related to Al usage) Employee/trade union involvement when new technologies are introduced Impact of Al/AM systems on working time and the right to disconnect Other topics Respecting the relevant privacy/data protection legislation Employee/trade union involvement in data protection Not specified 10,34% (3 CBAs) Al/AM tools used for monitoring and worker surveillance (e.g., software/devices to track physical or digital worker activity) 10,34% (3 CBAs) Use of AI/AM in recruitment process, work organization and worker’s assessment 3,45% (1 CBAs) 0 10 20 30 75,86% (22 CBAs) 62,07% (18 CBAs) 48,28% (14 CBAs) 48,28% (14 CBAs) 41,38% (12 CBAs) 37,93% (11 CBAs) 40 50 60 70 80 90 100 100%= all CBAs that refer to the introduction of Al and/or AM at work Source: WageIndicator Collective Agreements Database, accessed in November 2023. N=30 Table 3: Number of agreements that include clauses on the relevant topics and the years when the agreements commenced 2018 2019 2020 2021 2022 2023 Training for employees and/or management on new AI tools(including the risks related to Al usage) 1 2 3 5 6 5 Employee/trade union involvement when new technologies are introduced 2 3 3 2 6 2 Impact of Al/AM systems on working time and the right to disconnect 3 1 2 3 5 Other topics 2 2 2 1 4 3 Respecting the relevant privacy/data protection legislation 2 3 2 1 3 1 Employee/trade union involvement in data protection 2 2 2 3 2 Not specified 1 2 AI/AM tools used for monitoring and worker surveillance(e.g., software/devices to track physical or digital workers. 1 1 1 Use of AI/AM in recruitment process, work organization and workers assessment 1 Grand Total 3 4 4 5 7 6 Source: WageIndicator Collective Agreements Database, accessed in November 2023. N=30 Finally, Figure 15 shows when Artificial Intelligence(AI) and/ or Algorithmic Management(AM) at work was included in collective agreements in the different countries. 19

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 15: Artificial Intelligence(AI) and/or Algorithmic Management(AM) at work included in collective agreements in the different countries. Nr of CBAs referring to the introduction of AI or AM at work 6 5 4 Finland Germany Italy 3 Norway Spain 2 Switzerland United Kingdom 1 0 2018 2019 2020 2021 2022 2023 Source: WageIndicator Collective Agreements Database, accessed in November 2023. N=30 3.2 AI-RELATED CONTENT IN CBAS In this section, we highlight several CBAs that demonstrate good practice regarding rules and arrangements to govern the use of AI. These provisions may serve as a benchmark in collective bargaining in other cases/sectors/countries as well. The provisions summarised below originate(a) in the survey responses presented in Section 2, as well as in the content analysis of the 31 CBAs collected containing provisions gov­ erning the use of AI. Provisions contained in three CBAs are presented in greater detail, followed by an overview of addi­ tional CBAs with relevant AI-related stipulations. Box 1 Norway: Basic agreement for 2018–2021 between NHO and LO with amendments This CBA involves the Confederation of Norwegian Enter­ prise(NHO) and the Norwegian Confederation of Trade Unions(LO) and covers all respective associations, un­ ions, and enterprises. It addresses the opportunities and dilemmas of using AI at the workplace, emphasising ad­ herence to legal provisions. Employee privacy and dignity are paramount, requiring sound AI methods, involvement of employee representatives, and prevention of bias. The agreement references Chapter 9 of the Working Environ­ ment Act on control measures, requiring discussions with shop stewards in the case of important changes. Compa­ nies must keep employees informed through shop stew­ ards on plans and decisions regarding control measures, which can be based on technological, financial, safety, and health considerations. Control measures that are introduced must not go be­ yond the necessary scope, and must be factually justified in the individual company’s operations and needs. Em­ ployees or groups of employees must be treated equally when control measures are carried out.  ▶▶ ▶▶  Before measures are taken, company management and shop stewards individually and jointly act to ensure that employees receive information about the measures’ purpose and practical consequences, including how they are to be carried out, as well as the expected duration of the measures. To the extent that control measures address personal data, questions relating to the retention period, storage, deletion, etc., are discussed with the shop stewards and clarified in accordance with the Data Privacy Act and as­ sociated provisions. At regular intervals, the parties must evaluate the control measures implemented. If one of the parties to the individual company so de­ sires, a local agreement on the design must be sought as well as on implementation of the company’s control measures and the area of use for these. If no agreement is reached, each party is to bring its case before the main organisations, the Confederation of the Norwegian En­ terprise(NHD) and the Norwegian Confederation of Trade Unions(LO)(See Annexe for more details). Source: Survey response As to specific AI/AM-related content in CBAs, there is one example of good practice from Germany in 2020: the IBM Konzernbetriebsvereinbarung über die Einführung und den Einsatz von Systemen der Künstlichen Intelligenz / Artificial Intelligence. This CBA provides an extensive regulation of various AI-related topics, and the full text can be consulted here. 20

Annotation, coding and analysis of collective agreements’ contents The collective agreement for salaried employees in the ICT sector in Finland, concluded in 2023 and valid from 3 March 2023 until 31 March 2025 is particularly interest­ ing because it provides for the creation of a working group on the future of ICT work to also discuss AI-related topics. The parties are to establish a working group to continue the dialogue on the outlook of work in the ICT sector during pre­ vious agreement periods. The parties are also to encourage companies to continue this dialogue at the level of work­ places with a view to improving productivity, competitivity as well as staff well-being and competences. The following topics, among others, are to be reviewed dur­ ing the period of the Agreement: – the impact of digitalisation, robotics and AI on the work of salaried employees – changes in the demand for expertise caused by changes to working life and the recognition of future needs for expertise – maintenance of work ability, particularly in future cir­ cumstances and the central factors contributing to cop­ ing at work. The provisions of this CBA can be accessed at https://wagein­ dicator.fi/tyooikeus/tyoehtosopimustietokanta/the-collectiveagreement-for-salaried-employees-in-the-ict-sector-20232025-final In 2023, several CBAs also emerged in Spain that contain provisions on the use of AI. In most cases there is just a brief/general reference to AI. For example, in the health and safety section a clause has been added to take into account the evolution of technology, or in the trade union section a clause has been added on keeping the factory union in­ formed and consulted about new ways of working. One of the topics that is more commonly addressed in CBAs is digital rights at the workplace and the right to disconnect. A more detailed insight in CBAs stipulating digital rights is provided in Box 2. The Spanish CBA Convenio Colectivo de Empresa de Fundación Tekniker(2023–2024) is particularly interesting because of its provisions relating to the right to disconnect (see Box 3). Box 2 Spain: Agreement between Spanish Trade Unions and JUST EAT(2021) SIMA-FSP Fundación Servicio Interconfederal de Medi­ ación y Arbitraje F.S.P." is an Agreement from 2021 re­ sulting from a mediation procedure between Federación de Servicios CCOO, FSC- CCOO y FeSMC-UGT and TAKE­ AWAY EXPRESS SPAIN, S.L.(JUST EAT) Article 68, which is dedicated to digital rights, stip­ ulates inter alia that workers have the following digital rights at the workplace: a) Right to digital and work disconnection: Workers are to have the right to digital disconnection to guaran­ tee, outside of legal or conventionally established work time, respect for their rest and recuperation time, per­ mits and vacations, as well as their personal and ▶▶ ▶▶  family privacy. Consequently, the company is not to communicate with workers outside their working hours unless exceptional circumstances arise that justify such and/or to communicate the weekly work schedule to the delivery group. The signatory parties consider that digital disconnec­ tion is a right respect for which contributes to the health of workers by reducing, among other things, technological fatigue or stress, thus improving the work environment and the quality of work. (see Annex for more details) Source: https://tusalario.es/leys-laborales/base-de-datos-deconvenios-colectivos/convenio-takeaway-express-spain-s-l-justeat-2021-2023 Box 3 Spain: Convenio Colectivo de Empresa de Fundación Tekniker(2023–2024) This CBA includes a long article dedicated to the right to disconnect. This clause is considered to be the most extensive from among the 31 CBAs analysed. In compliance with the right to digital disconnection, this framework of action is established to ensure rest and recuperation time for all Tekniker staff, as well as respect for their personal and family privacy, regardless of the place or type of work day. This is made mandatory for Tekniker and all staff. 1. Scope of application: Tekniker guarantees all workers the right to digital disconnection once the workday has ended. All Tekniker staff have the right not to re­ spond to any communication, regardless of the me­ dium used(email, messaging apps, telephone, etc.), once their day has ended, unless the circumstances indicated in point 4 of this document come about. Digital disconnection is recognised and formalised as a right, although not as an obligation. Those people who have communications outside of the usual work­ ing day may do so freely, but they may not receive any response until another time during the working day. 2. Proper use of computer media: All Tekniker staff are committed to the appropriate use of computers and technological media made available to them and, among other aspects, ensure proper rest and recupera­ tion time by avoiding or minimising the use of said media for reasons relating to work outside the working day. Therefore, by virtue of the right to digital disconnec­ tion, Tekniker staff can disconnect the communication devices provided once their work day has ended, and they have no obligation to respond to messages or calls outside of working hours(except as required), as set out in clause 4 below. People with responsibility for managing a team must especially comply with the digital disconnection guide­ lines and are to refrain from requiring a response to possible communications sent to other people in the Center outside of working hours. Therefore, the re­ cipients of the communication will have the right not to respond to said communications made outside of work hours during rest and recuperation periods. ▶▶ 21

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors ▶▶  3. Place and day: The right to digital disconnection is understood in any workplace(including teleworking) and outside the usual working day or the alternative working day agreed upon in cases of teleworking or conciliation, including the vacation period, days for attending to personal matters, permits, sick leave, leaves of absence, or other matters of a similar nature. In cases of absence from work, all people have the ob­ ligation to leave a warning message by email(»absent from the office«) indicating the contact information of the colleague who will be able to respond in their absence, as well as the expected start and end dates of the period of absence. (See Annex for more details) Source: https://tusalario.es/leys-laborales/base-de-datos-deconvenios-colectivos/boletiofgipuzkoa-jueves-a-17-de-agostode-2023 22

Conclusions and lessons learned 4 CONCLUSIONS AND LESSONS LEARNED This report presents novel empirical evidence on the current state of the art in collective bargaining on AI and algorithmic management. The evidence reflects the views and experi­ ence of UNI Europa affiliates throughout the service sector in Europe, as well as the analysis of specific CBAs that contain AI-related stipulations and could serve as a benchmark or good practice for bargaining on AI-related worker control in other sectors and countries. 2 The survey shows that 42% of affiliates of UNI Europa are engaged in negotiations on various topics related to AI. Un­ ions are currently focusing on AI-related issues such as data protection and privacy for workers. The second most ne­ gotiated topics include the impact of AI on working hours, monitoring of worker activities, and automated scheduling of work shifts. On the other hand, the hiring process is the least common topic of negotiation among union members. The order of AI-related topics that have not yet been bar­ gained differs from those that have already been bargained. This order can be seen as a»wish list“ for future trade union negotiations and includes specific provisions in the collective bargaining agreement. The most requested topic relates to workers’ rights, specifically their right to challenge decisions made through automated decision-making, and their right to receive advice from an external data expert. Additionally, there is an intense desire for a right to information and con­ sultation on the use and evaluation of AI tools. Other vital topics include staff training on new AI tools, and automated background checks of social media profiles for job applicants. Out of the 90 responses, only 20% of trade unions reported having a collective agreement that addresses AI-related is­ sues at the organisation or sector level. This indicates that most trade unions(69%) do not have any collective bargain­ ing agreements relating to AI, and that 11% are unaware of any such agreements. The issues addressed most frequently in such agreements pertain to the protection of workers’ personal data and privacy. Provisions include restrictions on sharing workers’ personal data, collecting, processing, stor­ ing, and using their personal data, as well as complying with relevant privacy and data protection legislation. It appears that there are several reasons for the absence of AI-related stipulations in CBAs. Firstly, the AI topic has only recently become a topic of discussions and negotiations, hence there has not been enough time to compile a com­ prehensive CBA. Secondly, trade unions may not consider AI-related stipulations to be necessary, or they may have em­ ployer counterparts who are unwilling to address such issues, there may be insufficient resources to address these issues in depth or there may be other pressing concerns. Finally, the employer may have refused to engage in discussions on AI, and there may be a dearth of information on the subject. Above and beyond the CBAs, trade unions mainly rely on consultation to support the development of new legislation related to AI. Policy papers on AI and training / workshops that address risks associated with AI are also frequently used. Checklists or guidelines for introducing and using AI in the workplace have been utilised to a lesser extent. The topics that are neither being bargained nor for which there is a desire to bargain them indicate the secondary rel­ evance of AI-related issues for trade unions. On the other hand, it may indicate that AI-related topics are not seen as a threat to workers to date or that trade unions have limited information on these matters. The least relevant topics in AI-related bargaining include access to employee emails and messages, systems for generating automatic responses(e.g. chatbots), and voice analytics(e.g. in contact centres). The content analysis of CBAs shows that AI has indeed been emerging as a relevant bargaining topic, especially in very recent years, with CBAs increasingly containing AI-related stipulations. While some of these are just general references to the use of technology, the research was able to identify several CBAs that can serve as examples of a more detailed governance of the right to disconnect, digital rights of work­ ers at the workplace, information-sharing and business con­ trol. 2 The second workstream of FES and UNI Europa’s effort to explore AI and Algorithmic management is a study on frequently used AI systems in in European services sectors: prevalence, functions, and a guide for negotiators prepared by Steve Rolf, University of Sussex Business School In sum, bargaining on AI is an evolving phenomenon and is not yet as widespread as provisions relating to other ele­ ments of working conditions. However, trade unions have already started addressing AI and algorithmic management. Given the increasing use of technology at the workplace and 23

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors current opinions of trade unions regarding bargaining on AI-related stipulations, it can be expected that collective bar­ gaining on AI will further gain in relevance and there will be more CBAs in the future. Providing good practices on CBAs that already govern the use of AI can serve as a guardrail in the development of this bargaining field while encourag­ ing informed decisions by parties engaging in bargaining in many countries. REFERENCES Aloisi, A.,& De Stefano, V.(2021). Essential jobs, remote work and digi­ tal surveillance: addressing the COVID-19 pandemic panopticon. International Labour Review, 1–31. doi:10.1111/ilr.12219 Collins, P.,& Atkinson, J.(2023). Worker voice and algorithmic management in post-Brexit Britain. Transfer: European Review of Labour and Research, 29(1), 37–52. doi:10.1177/10242589221143068 De Stefano, V.,& Taes, S.(2023). Algorithmic management and collective bargaining. Transfer: European Review of Labour and Research, 29(1), 21–36. doi:10.1177/10242589221141055 Eurofound.(2021). Riders’ law(Initiative), Record number 2449. Platform Economy Database. Dublin. Retrieved from https://apps.eurofound.europa. eu/platformeconomydb/riders-law-105142 High-Level Expert Group on Artificial Intelligence.(2019). A definition of AI: Main capabilities and disciplines. European Commission. Retrieved from https://ec.europa.eu/futurium/en/system/files/ged/ai_hleg_ definition_of_ai_18_december_1.pdf Krzywdzinski, M, Gerst, D., Butollo, F.(2023). Promoting human-­ centred AI in the workplace. Trade unions and their strategies for regulating the use of AI in Germany, Transfer: European Review of Labour and Research, vol. 29(1), pages 53-70, February. DOI: 10.1177/ 10242589221142273 Mateescu, A.,& Nguyen, A.(2019). EXPLAINER: Algorithmic Manage­ ment in the Workplace. Data& Society. Retrieved from https://datasociety. net/wp-content/uploads/2019/02/DS_Algorithmic_Management_Explainer. pdf Nordic Financial Unions.(2022). Artificial intelligence in the Nordic finance sector: risks, challenges, and opportunities for finance employees. Retrieved from https://nordicfinancialunions.org/wp-content/uploads/ NFU-Position-Paper-Artificial-intelligence-in-the-Nordic-finance-sector.pdf Prospect.(2021). Digital technology: Guide for union representatives. Retrieved from https://library.prospect.org.uk/download/2022/00121 RMG& CWU.(2020, December 18). KEY PRINCIPLES FRAMEWORK AGREEMENT. Rodríguez, L.(2022, January 13). First collective agreement for platform workers in Spain. Social Europe. Retrieved from https://www.socialeurope. eu/first-agreement-for-platform-workers-in-spain Schwartz, D. E., Jeffers, A. V. and Safko, E. D.(2023). AI and the Workplace: Employment Considerations, Skadden Insights. Retrieved from https://www.skadden.com/insights/ publications/2023/06/quarterlyinsights/ai-and-the-workplace Telefónica.(2018, November 23). Telefónica recognises the right of its employees to digitally disconnect. Retrieved from https://www.telefonica. com/en/communication-room/press-room/telefonica-recognises-the-rightof-its-employees-to-digitally-disconnect/ Todolí-Signes, A.(2021). Spanish riders law and the right to be informed about the algorithm. European Labour Law Journal, 12(3), 399–402. doi:10.1177/20319525211038327 UNI Global Union.(2023, May 12). ALGORITHMIC MANAGEMENT: OPPORTUNITIES FOR COLLECTIVE ACTION. Retrieved from https:// uniglobalunion.org/wp-content/uploads/Algorithmic-ManagementOpportunities-for-Collective-Action.pdf 24

Annexes ANNEXES 1. SAMPLE OF ONLINE SURVEY RESPONSES AMONG UNI EUROPA AFFILIATES Table 4: Number of responses by country Country France Belgium Denmark Italy Norway Great Britain Ireland Romania Austria Netherlands Spain Finland Germany Sweden Iceland Croatia Czech Republic Hungary Luxemburg Switzerland Turkey Bulgaria % 13.51% 12.84% 6.08% 6.08% 6.08% 5.41% 4.73% 4.73% 4.05% 4.05% 4.05% 3.38% 3.38% 2.70% 2.03% 1.35% 1.35% 1.35% 1.35% 1.35% 1.35% 0.68% Number of answers 20 19 9 9 9 8 7 7 6 6 6 5 5 4 3 2 2 2 2 2 2 1 ▶▶ Country Estonia Latvia Malta Poland Portugal Slovakia Israel Cyprus Albania Greece Lithuania Slovenia Kosovo Bosnia-Hercegovina Serbia Moldova Monaco San Marino Ukraine Russian Federation Belarus Kingdom of Bahrain % 0.68% 0.68% 0.68% 0.68% 0.68% 0.68% 0.68% 0.68% 0.68% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 99.35% Number of answers 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 148 25

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Table 5: The number of responses in the survey by sector(N=148, as a%) Sector ICTS sector(Information, Technology and Services) Finance Post and logistics Commerce Other Media and entertainment(prints, TV, radio, internet broadcasting, etc.) Agency work Cleaning and security Graphical and packaging Care(nursing and home care sector) Tourism Gaming(casinos, lotteries, betting shops, and other gambling enterprises) Hair and beauty Total Q5: In which sector do you conduct collective bargaining? Number of responses 36 24 23 17 14 12 5 5 5 3 2 1 1 148 % 24.3 16.2 15.5 11.5 9.5 8.1 3.4 3.4 3.4 2.0 1.4 0.7 0.7 100.0 2. ADDITIONAL FINDINGS FROM THE ONLINE SURVEY OF THE UNI EUROPA AFFILIATES Figure 16: Position of the respondent in the trade union (N=148, as a%) 60% 49 50% Figure 17: Involvement of the trade union of the respondent in the collective agreement(N=148, as a%) 80% 71 70% 40% 60% 30% 28 50% 20% 13 11 10% 0% National trade union officer Local/regional trade union officer Sectoral trade union officer Q3: What is your position in the trade union? Company shop steward/delegate Trade unionists could reveal the name of the employer or company that they are bargaining with. Not all stated their counterparts in bargaining. The majority provided the name of the company, however. The table below presents exam­ ples of social partners involved in the bargaining process. 40% 30% 22 20% 10% 5 3 0% I directly I don't directly I am not at all Other negotiate negotiate involved in (please specify) collective collective agreements bargaining Q4: How are you involved in collective bargaining? 26

Annexes Table 6: Trade unions and their bargaining counterparts by country Country Italy Belgium Netherlands Belgium France Denmark Norway Romania Denmark France France Sweden Norway France Germany Bulgaria Belgium Austria Switzerland Ireland France Netherlands Spain Trade union Slc Cgil CSC-ACV Transcom Culture FNV ACV Puls FO Communication Finansforbundet Fellesforbundet Federatia FAIR-MEDIASIND HK Privat SYNEE FO Banques The Commercial Workers Union of Sweden, Handels HK i Norge FEC FO ver.di Trade union federation of communications ACOD – CGSP GPF – Gewerkschaft der Post und Fernmeldebediensteten syndicom – Gewerkschaft Medien und Kommunikation Communications Workers’ Union(CWU) Ireland Fo FNV UGT Employer / company CCNL Telecomunicazioni In different Trade Unions. IT Hardware sector E.a. Ricoh, Xerox, Canon, Fujitsu, HPE, Konica Minolta, Acer, KyoceraToshiba,a, A Commerce LA POSTE Finansforbundet The national employers federation FAIR-MEDIASIND Dansk Industri, Dansk Erhverv and other major national employer organisations Pôle emploi Up Coop Sectoral level bargaing and company level Virke and Norges Hovedorganisasjon(NHO) Secteur Banques et Sociétés Financières IBM, Kyndryl, Tui Infotech, T-Systems… BG Post PLC bpost Österreichische Post AG Swisscom Eir, Vodafone, Circet Bpmed PostNL Telefónica Figure 18: Trade unions and their bargaining counterparts by country 80% 69 70% 60% 50% 40% 30% 20 20% 11 10% 0% Yes No Don’t know Q13: Is there a collective bargaining agreement on AI-related topics in your organisation / sector? 27

FRIEDRICH-EBERT-STIFTUNG – Collective Bargaining Practices on AI and Algorithmic Management in European Services sectors Figure 19: Level at which the collective bargaining agreement was concluded?(N=16, as a%) Sector level(bargaining on behalf of the whole sector) Multi-employer level(bargaining with representatives of several employers) Company/organization level(bargaining in a single employer with multiple workplaces, e.g. a bank with... Other(please specify) Workplace level(bargaining only at one workplace) 0 Q14: At what level has the collective bargaining agreement been concluded? 25 25 25 19 6 5 10 15 20 25 30 Table 7: Examples of other CBAs that contain AI-related provisions (as reported by survey respondents) – Convenio colectivo BANCA XXIV – Collective agreement for Telenor between the Confederation of Norwegian Enterprises(NHO) on the one side and the Norwegian Confederation of Trade Unions(LO) and the Nor­ wegian Electricians and IT Workers Union(EL og IT Forbundet) on the other – »Gesamtarbeitsvertrag Swisscom: – Swisscom achtet das Recht der Mitarbeitenden, während ihrer Freizeit nicht erreichbar zu sein.« – V ACUERDO PARA EL EMPLEO Y LA NEGOCIACIÓN COLEC­ TIVA(V AENC) – TAKEAWAY EXPRESS SPAIN, S.L.(JUST EAT) – Openreach Service Delivery Transformation – GPS Technology Code of Practice – As mentioned, we do not have any specific text on AI, but a broader text on involvement when introducing new tools, and texts regarding involvement and restrictions about sharing and using data regarding employees work, etc. – Ccnl Telecomunicazioni – Handelsoverenskomsten LO/NHO Service og Handel – Landsoverenskomsten LO/Virke" – Settore TLC – HST initiates Collective Bargaining in Combis 3. AI-RELATED CONTENT IN CBAS Box 1 Norway(continued) In addition to the above provisions, this CBA also con­ tains an additional Framework Agreement on Technological Development and Computer-Based Systems (IV). The Main Organization of Business and the National Organization Norway agree that this framework agree­ ment should be consulted when designing, introducing, and using technology and computer-based systems. The framework agreement is based on and does not imply any limitation of the Main Agreement’s Part A. Moreover, the parties agree that the agreement’s provisions must be ap­ propriately put into practice depending on the company’s size without weakening the intentions of the agreement. In this agreement, the term technology includes tech­ nology relating to production(including automation), ad­ ministration and management systems. The agreement includes technology and systems used by planning and implementation of the work and systems for storing and using personal data that apply to em­ ployees at the individual company, cf. Main Agreement § 9–11. It is further provided that the agreement applies in addition to computer-based systems to significant tech­ nological changes and cases in which such changes are important to employees and their working conditions. When a company plans and adopts a computer sys­ tem, it must the area of use is clearly defined. Other uses of systems could only happen after discussions with the employees’ shop stewards. To a reasonable extent, the shop steward must use the company’s expertise in consultation with company man­ agement. If necessary, the employee’s shop steward – in agreement with the company management and through its main organisation – could consult external expertise in the area. The costs of such expert help are covered by the company unless otherwise agreed in advance.  ▶▶ 28

Annexes ▶▶  Before the company makes a final decision about system and/or technology choice within the agreement area, the parties to it should discuss how the employees can participate in developing and/or participating in se­ lecting such technology. Training and retraining needs must be clarified. It shall particular emphasis be placed on the development of the individual’s professional skills and job content. Box 2 Spain(continued) b) Right to privacy and the use of digital devices in the workplace: Digital devices made available to workers by the Company are exclusively for professional purposes, prohibiting personal use of the same. The Company may access content derived from the use of digital media provided to the workers to control compliance with obligations laid down in labour or stat­ utory regulations and guarantee the integrity of said de­ vices. The control of the use of technological tools is to be carried out in all cases respecting the principle of propor­ tionality, as well as the dignity of workers and with all of this being subject to the right of workers to the protection of their privacy in the use of digital devices made available to them by the Company. c) Right to privacy against using video devices, sur­ veillance and sound recording at the workplace in the case of cameras or video camera systems as well as video recording systems and sound systems for the exercise of personnel control being installed. f) Information on digital work tools: In addition to the above, both the Algorithm Commission and each em­ ployee of the Company. You will be adequately informed of all digital tools that exist in your environment, work, and possible associated health risks. In the same way, when workers become involved in a conversation, they must be informed in advance whether they are talking to a chatbot or a person. All conversa­ tions will be recorded and kept accessible for a period of three months for working people and will later be de­ leted. These conversations carried out by a chatbot can­ not be used to sanction a person. 4. Promoting good practices: Tekniker will carry out awareness-raising actions about the right to digital dis­ connection, informing and/or training people about the necessary protection of this right, while taking into ac­ count the circumstances, both work-related and personal, and to this end the information they need will be made available to promote their understanding and subsequent application. All departments must promote responsible practice with regard to information and communication technologies on the teams under their charge. In response to this possibility of needing urgent at­ tention, and even recognising the right to digital discon­ nection, it is considered necessary that certain positions remain reachable and accessible even during rest and re­ cuperation time. These positions are:  ▶▶ ▶▶  – Director of People and Organisation, – Economic-Financial and General Services Director, – Information Manager and IT Infrastructure Manager. – Maintenance staff. In cases where some of these extraordinary causes justify the activity of any of the indicated persons, interrupting their rest and recuperation time, this time is deemed to be effective working time, and its provisions shall be in ac­ cordance with the provisions of the Tekniker Agreement. 5. Causes of exception: Recognising the right to dig­ ital disconnection of all people, including management, the existence of some activities or the possibility of the appearance of critical circumstances that justify the need to locate and respond to some people cannot be ignored. In particular, those situations in which failure to afford urgent attention could give rise to serious harm to busi­ ness, whose sudden urgency clearly requires an immedi­ ate response, or those situations that could pose a serious risk to people, equipment or facilities at the Centre, and those cases in which it is necessary to respond to urgent contact requests from public authorities, emergency ser­ vices or private security hired by Tekniker. 29

IMPRINT ABOUT THE AUTHORS IMPRINT Marta Kahancová, PhD., is the co-founder and manag­ ing director of the Central European Labour Studies Institute (CELSI) and an associate professor in public policy at Come­ nius University, Faculty of Social and Economic Sciences, Bratislava, Slovakia. Barbora Holubová, PhD., is a senior researcher at CELSI in Bratislava. Simona Brunnerová is a junior researcher at CELSI in Bra­ tislava. Katarína Lukáčová is a researcher at CELSI in Bratislava. Published by Friedrich-Ebert-Stiftung| Competence Centre on the Future of Work| Cours Saint Michel 30e| 1040 Brussels| Belgium Dr. Tobias Mörschel, Head of Friedrich-Ebert-Stiftung Competence Centre on the Future of Work Responsible Contact: Oliver Philipp oliver.philipp@fes.de For more information about the Competence Centre on the Future of Work, see: https://futureofwork.fes.de/ Daniela Ceccon is Director Data at WageIndicator Foun­ dation. Gabriele Medas is the Collective Agreements Database manager of the WageIndicator Foundation. Design: Petra Strauch/Barbora Novotna The views expressed in this publication are not necessarily those of the Friedrich-Ebert-Stiftung. Commercial use of media published by the Friedrich-Ebert-Stiftung(FES) is not permitted without the written consent of the FES. Publications by the FES may not be used for electioneering purposes. ABOUT CELSI The Central European Labour Studies Institute(CELSI) is a non-profit research institute based in Bratislava, Slovakia. It fosters multidisciplinary research on the functioning of la­ bour markets and institutions, work and organisations, busi­ ness and society, ethnicity and migration in the economic, social, and political life of modern societies. CELSI strives to make a contribution to cutting-edge international scholarly discourse. In its applied research and policy analysis, CELSI promotes a focus on Central and Eastern European coun­ tries. ISBN: 978-3-98628-550-0 © 2024 ABOUT WAGEINDICATOR WageIndicator Foundation is a global, independent, non­ profit organisation that collects, analyses and shares informa­ tion on Minimum Wages, Cost of Living and Living Wages, Salaries and Wages, Labour Laws, Gig and Platform Work, and Collective Agreements across 206 countries worldwide. Their mission is to improve labour market transparency to ensure workers, employers, trade unions and other stake­ holders can make informed decisions. www.fes.de/bibliothek/fes-publikationen