Employee Toolkit
Algorithmic management systems are increasingly being used to instruct, evaluate, discipline, and reward workers. While this phenomenon is typically believed to have its origins in the gig economy, it is important to acknowledge that the use of technology to evaluate and manage workers has much earlier roots. In the late 1800s, Frederick Taylor was arguing for a 'scientific management approach' to labour to maximise efficiency in the workplace. However, what is new is the scale, ubiquity, and impact of algorithmic management practices. Now, workers may experience algorithmic management from the very beginning of the hiring process all the way to the end of the employment relationship.
In our toolkit for employees, we provide a summary of some of the legal rights and protections available to workers subjected to algorithmic management. There is no single legislative instrument designed to ensure that algorithmic management is lawful, transparent, accurate, and fair. Nonetheless, existing laws such as anti-discrimination laws, data protection, and newer labour and technology laws apply, and all offer some protections. This toolkit also notes some gaps and limitations of these existing laws when applied to algorithmic management.
This toolkit summarises and extracts key information from over five years of research conducted by the iManage team, led by Professor Jeremias Adams-Prassl. Our analysis focuses on the UK and EU contexts, but may be applicable to other jurisdictions with similar legal landscapes or be drawn upon for comparative analysis. The toolkit only offers a snapshot of our research, with the full published papers linked where applicable and a further research list at the bottom of the page.
What are the existing rights and protections that apply to algorithmic management? What are their limits?
People may think that because AI is new, the law doesn’t apply. But that’s a very convenient myth. The starting point must be that existing rules – including those on discrimination, privacy and fairness – still hold. — Professor Jeremias Adams-Prassl
Data protection law and algorithmic management
Data protection has been one of the areas of law most engaged in the context of algorithmic management, owing to the extensive personal data processing to which workers are subjected in many work contexts. In 2025, for example, an OECD survey of 6,000 managers across six OECD countries found that 90% of managers surveyed reported that their firms were using at least one algorithmic management tool. The analysis below focuses on existing provisions in the EU and UK General Data Protection Regulation (GDPR) that could offer some protection against the harms and risks arising from algorithmic management. In this context, 'data subjects' refer to workers.
Articles 12-14 of the EU and UK GDPR stipulate how and when workers must be informed with respect to the processing of their personal data. They apply without being invoked by the worker. In the context of algorithmic management, the right to be informed provides one instrument to counterbalance the informational power asymmetry in the employee-employer relationship. These rights also enable workers, as data subjects, to invoke other data subject rights—e.g., rectification of inaccurate personal data, deletion of personal data, and data portability.
Key elements include:
- The right to be informed (Article 12-14 UK GDPR; Article 12-14 EU GDPR) applies regardless of whether personal data is collected from the worker or from other sources – for instance, this may prove helpful in the context of hiring systems which draw upon data from individuals' online profiles or other databases to inform decision-making (e.g., see here)
- It is not limited to personal data processing, but the right to be informed also extends to information on processing operations and information on workers' rights
- The right to be informed is part of the overarching transparency requirement embedded in data protection law. Therefore, information must be provided to the worker 'in a concise, transparent, intelligible and easily accessible form, using clear and plain language' (Article 12(1) EU GDPR; Article 12(1) UK GDPR)
But there are gaps and limitations...
For instance, employers may attempt to reject data access requests that they consider ‘manifestly unfounded or excessive’ (Article 12(5) EU GDPR; Article 12(5) UK GDPR). Notably, however, the employer ‘bear[s] the burden of demonstrating the ‘manifestly unfounded or excessive character of the request’, meaning that if the employer rejects a request using this reasoning, but the worker genuinely made the request in good faith, the worker may consider filing a complaint to the data protection authority.
Employers can also ask workers who request access to their data to specify the data they wish to receive or the processing activities about which they wish to be informed (Recital 63 UK GDPR; Recital 63 EU GDPR). This can pose a problem if the employer is processing large amounts of worker data and the worker does not know what to ask for. Workers in this situation may consider (a) asking the employer for a concise overview of the personal data being processed and the purposes for which it is being processed; and/or (b) consulting a worker representative (e.g., trade union or works council).
Research about the right to be informed and worker data rights broadly:
- Halefom Abraha, 'Regulating algorithmic employment decisions through data protection law', European Labour Law Journal 14(2), 2023
- Halefom Abraha, 'Navigating workers' data rights in the digital age', International Labour Organization Working Paper 149, September 2025
The ICO (Information Commissioner's Office), the UK’s data protection authority, maintains extensive data protection guidance and checklists for employers processing workers’ personal data. If you are a worker and you suspect your employer may be processing your personal data in violation of your rights under GDPR, these resources may help you determine if this is the case. (Note for EU readers: As of November 2026, these are being updated after post-Brexit changes to the UK GDPR.)
The data subject’s right of access to personal data undergoing processing is set out mainly in Article 15 of the GDPR.
Key elements include:
- The right of access includes a right to obtain: (i) information about whether personal data are being processed; (ii) further information about that processing; and a copy of all this personal data.
- Personal data includes any information ‘relating to’ an identifiable person. Information can ‘relate to’ a person through its ‘content’ (i.e., by being ‘about’ that person) or by the purpose or effect of the processing.
- For example, a record of phone calls made from an office telephone may not be inherently ‘about’ a person, but it may nonetheless become ‘their’ personal data if the record of calls is used to investigate potentially fraudulent workplace behaviour, and the person comes under suspicion because the phone is in their office (see Opinion 4/2007 of the Article 29 Working Party ‘On the concept of personal data’ and C-434/16 Nowak v. Data Protection Commissioner).
- Data provided by third parties or generated by internal systems, such as customer reviews of worker performance or automatically generated scores, are the personal data of the person being scored or reviewed.
- Article 15 EU and UK GDPR provides an ex post right of access to data about automated decision-making (including the existence of automated decision-making, the logic involved, and the envisaged consequences), but it applies only when invoked by the worker (see Dun & Bradstreet)
- Workers can use the right of access to counterbalance information asymmetries in algorithmic management, exercise other rights (e.g., safeguards listed in Article 22 EU GDPR; see Dun & Bradstreet), and voice their concerns to their employer and trade unions.
- While the EU nor UK GDPR does not precisely define what constitutes 'meaningful information about the logic involved', it should be interpreted in line with the underlying aim of the right of access and principle of transparency (Article 12 UK GDPR; Article 12 EU GDPR, also see Dun & Bradstreet)
But there are gaps and limitations:
The right of access is also subject to carve-outs and limitations. Employers can refuse or limit the right of access if this is necessary to protect the rights and freedoms of others and intellectual property protections. However, the logics or algorithms or processes used to make significant decisions about workers (e.g., in recruiting, work organisation, performance evaluation, discipline, etc.) are typically not going to be genuinely competitively sensitive, so arguments that employers (or technology providers) don’t need to provide explanations or access to data on grounds of trade secrets or intellectual property should be scrutinised. Furthermore, where there are genuine concerns, this cannot result in the complete failure to consult or provide information (explanations/access to data) (Recital 63 UK GDPR; Recital 63 EU GDPR; Dun & Bradstreet). Where necessary, the conflict should be adjudicated by a competent supervisory authority.
Data controllers do have the right to refuse to provide access to data if the request is 'manifestly unfounded or excessive' (Article 12(5) EU GDPR; Article 12(5) UK GDPR). However, the burden of proof sits with the controller and where data subjects make a good faith request and it is refused by the controller, the data subject has recourse to the data protection authority. Furthermore, the right of access is unconditional and does not depend on the purpose for which the data subject intends to use the copy of the data; there is no such thing as ‘abuse’ of the right of access (see Li and Toh, 2022).
Data Protection Impact Assessments (DPIA) are limited in scope, but may provide protections to workers in upstream decision-making on the impacts of algorithmic management. Data controllers are required to carry out a DPIA when processing is 'likely to result in a high risk to the rights and freedoms of individuals (Article 35(1) UK GDPR; Article 35(1) EU GDPR).
Key elements include:
- DPIAs must include 'an assessment of the risks to the rights and freedoms of data subjects' as well as the 'measures envisaged to address' these rights (Article 35(7) EU GDPR; Article 35(7) UK GDPR).
- The majority of national data protection agencies in EU member states have explicitly identified that employee monitoring requires a DPIA (see here)
- In certain instances, data controllers must also involve workers, their representatives, and potentially the data protection authority (e.g., in the UK, the Information Commissioner's Office) prior to processing data, especially when there is a 'high risk' to workers' rights that the employer cannot mitigate (Article 36 UK GDPR; Article 36 EU GDPR)
But gaps and limitations also exist:
The obligations to carry out DPIAs in high-risk contexts typically fall short in terms of inclusivity of algorithmic management systems within the scope of assessment. Halefom Abraha provides an example of a system which processes anonymised data to inform supply chain efficiency which leads to an entire team being relocated which would fall outside the scope of a DPIA. In practice, some companies in the logistics industry use algorithmic systems to predict (through a score) warehouses’ likelihood of unionisation. The companies may claim these systems do not process ‘personal data’, perhaps on the theory that this exempts them from data protection obligations, including impact assessments.
Though they have uses, DPIAs have been criticised by some analysts as 'cosmetic audits, vague checklists and courtesy toolkits' that often have little practical impact on how algorithmic management systems are deployed. There is no obligation to publish the results of a DPIA (even to workers and representatives), meaning workers may have limited opportunity to voice concerns about shortcomings in the assessment.
Data protection laws also allow for more specific rules on employee data processing, particularly to 'ensure the protection of the rights and freedoms' of workers (Article 88(1) EU GDPR). Note: this section is only applicable in the EU, the relevant provision was removed from the domestic UK framework post-Brexit.
Key elements include:
- Provides protection of workers' 'human dignity, legitimate interests and fundamental rights' which may be threatened by the use of algorithmic management in the employment context (Article 88(2) EU GDPR)
- It requires that the substantive content of national rules at least stick to the minimum requirements of the EU GDPR, or provide a higher degree of protection to employees
- This provision can be used to strike down Member States' laws on employee data processing as incompatible with the EU GDPR and thus no longer applicable (see Hauptpersonalrat der Lehrerinnen and read more here)
- Member States also have an obligation to notify the Commission of their respective legal provisions on specific rules that they adopt in the context of employee data processing (Article 88(3) EU GDPR)
Discrimination law and algorithmic management
Subject to careful judicial development in the UK context, Adams-Prassl et al argue that existing norms in discrimination law will be able to tackle many supposedly novel legal questions relating to biases and algorithmic management. In spite of potential difficulties in liability attribution in complex, dispersed technology supply chains, discrimination law, developed in the common law courts, holds significant promise in tackling key mechanisms of algorithmic bias. In fact, the digitalisation of previously (often opaque) human decision-making processes may even render discriminatory criteria and outcomes more traceable and quantifiable.
Direct discrimination may apply in the context of algorithmic management when individuals are subjected to less favourable treatment on the grounds of, or because of, a protected characteristic (Section 13, UK Equality Act 2010). Protected characteristics (outlined in Section 4, UK Equality Act 2010) include (to name a few) age, disability, race, gender reassignment, sex and religion or belief. Salient examples show automated systems making decisions based directly on protected characteristics, without any proxies (see e.g., credit scoring in Finland). A finding of direct discrimination is subject to very few exceptions and affords the decision-maker very little scope for justification (unlike indirect discrimination, see Section 19, UK Equality Act 2010 and 'self-justifying feedback loops' in Adams-Prassl et al).
Key elements include:
- Direct discrimination has long applied to human decisions, and there is no reason that an inherently discriminatory criterion is any less discriminatory merely because it is applied by a computer system.
- Direct discrimination may even apply to criteria applied by algorithmic systems which are not a protected characteristic but a proxy for it (see Lee v Ashers).
- Direct discrimination may even apply when there is not a 'perfect proxy' for the protected characteristic (in the UK see R v Birmingham City Council; Chief Constable of West Midlands Police and others v Harrod and others, and EU context, see WABE)
- It may also apply when an algorithm identifies a set of factors which, when applied together, constitute a perfect proxy for a protected characteristic – even if no human would have realised that this result would occur.
- For a decision-maker to argue against the directly discriminatory nature of their algorithmic management system, they will have to prove that the algorithm's less favourable output was not because of a protected characteristic (or proxy of it) (see Lee v Ashers). This may be challenging if the workings of the algorithm are inscrutable.
But there are gaps and limitations:
In order to demonstrate that a claimant was treated less favourably by an algorithmic management system than an actual or hypothetical comparator, claimants may need to adduce statistical evidence to demonstrate unfavourable treatment (see Rihal v Ealing LBC). Challenges may arise to provide such evidence because of a combination of technical complexity and algorithmic opacity (read more here).
Platform worker protections and algorithmic management
Many digital labour platforms (e.g., food delivery, ride-hailing, and online freelancing) use fully or partially automated systems to evaluate applications for access to work, assign work, and evaluate work and persons performing work. The EU Platform Work Directive (PWD) was passed by the EU to improve working conditions in platform work (see Article 1, PWD), including protections for workers subjected to algorithmic management. Indeed, following the enactment of the PWD, platform workers, long unprotected by the law, now find themselves with some of the strongest protections against intrusive and extractive algorithmic management practices.
Transparency in the PWD focuses on empowering workers with knowledge about the uses and functioning of automated monitoring and decision-making systems.
Key elements include:
- Digital labour platforms are required to, by the first working day, provide workers with information concerning algorithmic management systems, prior to the introduction of changes which affect their working conditions which must be accessible and clear for workers to understand (Article 9 PWD)
- The 'main parameters' of automated monitoring systems and automated decision-making or support systems must be declared, with their importance and the ways that platform workers' personal data influences the decisions (Article 9(1)(b)(iii) PWD)
- A written statement on grounds for certain significant decisions made by an algorithmic management system must also be disclosed (e.g., refusal of payment; account suspension, restriction or termination) (Article 11(1) PWD)
- Following a decision taken by or supported by an algorithmic management system, platform workers are entitled to ‘access to a contact person designated by the digital labour platform to discuss and to clarify the facts, circumstances and reasons having led to the decision’ (Article 11(1) PWD)
Scholarship on algorithmic transparency in the Platform Work Directive:
- Earlier research from the iManage project on the algorithmic management rules in the Platform Work Directive:
Michael Veale, Six Silberman, Reuben Binns, 'Fortifying the algorithmic management provisions in the proposed Platform Work Directive', European Labour Law Journal 14(2), 2023
From outside the iManage project: Ilda Durri, Charalampos Stylogiannis and Mathias Wouters, 'The [EU] Platform Work Directive: A milestone or a nothingburger?' In S. Sabato et al., eds., Social Policy in the European Union: State of Play 2024, European Trade Union Institute & European Social Observatory (OSE).
'The Platform Work Directive cannot be characterised as a nothingburger', Durri et al. write after a careful review of the final version of the Directive.
The PWD includes numerous human oversight protections with respect to the ex post review of decisions, granting workers the right to human explanations of decisions and the ability to overturn decisions in certain circumstances. These are intended to "create accountability around the functioning of algorithmic systems that could affect thousands of workers, as well as to support individual justice around specific decisions" (Veale, Silberman and Binns, 2023).
Key elements include:
- For decisions significantly affecting working conditions, digital labour platforms are obliged to have a contact person with competence, training, and authority to discuss and clarify the facts, circumstances, and reasons that led to a given automated decision (Article 11(1) PWD).
- Decisions are also required to be rectified, and where rectification is not possible, workers should be offered compensation where the decision infringes the worker's rights, e.g., labour rights or non-discrimination (Article 11(3) PWD).
But there are gaps and limits:
The protections surrounding the overturning and compensation of automated decisions impacting workers' rights will often rest on national labour standards around those decisions—these can differ depending on a worker's status (e.g., the scope of unfair dismissal legislation differs widely across the EU) (read more here).
Article 13 of the PWD requires that Member States ‘ensure information and consultation of platform workers’ representatives on decisions likely to lead to the introduction of or substantial changes in the use of automated monitoring and decision-making systems.
Key elements include:
- The PWD acknowledges two scenarios for the fulfilment of digital labour platforms’ information and consultation obligations: either with platform workers’ representatives, or ‘where there are no such representatives, of the platform workers concerned’
But there are gaps and limits:
There is little guidance indicating how the process of information and consultation might proceed without representatives and only with platform workers. For instance, digital labour platforms may not have the resources, competence, or willingness to facilitate consultation with workers in different countries using multilingual processes (read more here). These provisions only apply to ‘platform workers’ meaning that legitimately self-employed persons performing platform work without an employment contract/relationship fall outside the scope of involvement in these information consultation measures (read more here).
Article 20 of the PWD also includes an obligation on digital labour platforms to 'design in, through the platform's digital infrastructure or similarly effective means, the possibility for persons performing platform work to contact and communicate with each other and through representatives'.
Key elements include:
- Informal discussion can lead to improvement in working conditions, e.g., by drawing attention to technical problems experienced by many those performing platform work
- The communication provisions will influence how and if information and consultation provisions are effectively realised in practice, e.g., especially when workers exist across borders
But there are gaps and limits:
If the communication channels provided by digital labour platforms are inaccessible, poorly organised, or technically unreliable, then this will pose challenges for workers to consistently ‘make collective sense’ of information they are receiving and to effectively be consulted (Veale, Silberman and Binns, 2023). Digital labour platforms may have to collect information that relates to private discussions from these communication channels (acknowledged in Recital 62, PWD). This may be invasive of workers' privacy if messages sit on personal devices but also may be practically challenging for employers to gain control of these devices without mobile device management policies which require users to surrender control of their devices (Veale, Silberman and Binns, 2023).
AI law and algorithmic management
AI systems are increasingly used in workplaces as part of algorithmic management systems (see here). In the EU AI Act, 'AI' is defined as 'a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments (see Article 3(1) EU AI Act). Algorithmic management can include both ‘generative AI’ technologies (e.g., chatbots) but also machine learning algorithms used to make rule-based decisions (e.g., in the context of CV screening) in the context of algorithmic management.
In the context of algorithmic management, the EU AI Act's 'high risk' AI provisions (Article 6, EU AI Act) are the most applicable, and subject these systems to risk management and design requirements.
Key elements include:
- The Act classifies employment as a high-risk area, and specifically pinpoints algorithmic management tools as within the scope of risk mitigation and design requirements for “high-risk AI systems” (Annex III, EU AI Act)
- Most of the high-risk AI obligations fall to the AI provider (the person/body that develops or markets the AI), rather than its user (the person/body who uses the system) (Article 16, EU AI Act)
- Every “high-risk” AI system placed on the market should be accompanied by a “conformity assessment” confirming that the requisite AI standards have been met (Article 16(f), EU AI Act; Article 43, EU AI Act)
- High-risk systems should be built to be "sufficiently transparent to enable deployers to interpret a system's output and use it appropriate" (Article 13, EU AI Act)
- Article 60 provides a publicly accessible “database” containing information about high-risk AI systems, including the ‘instructions for use’ for high-risk algorithmic management tools (Article 60, EU AI Act)
But there are gaps and remaining challenges:
In practice, while many algorithmic management systems will fall within the scope of 'high-risk' AI systems, only few will be prohibited under Article 5 — even if they pose significant harm to workers (Kelly-Lyth, 2021). Conformity assessments will also be self-assessments, meaning that technology companies will be 'marking their own homework' with respect to their compliance with the EU AI Act (Kelly-Lyth, 2021). Transparency obligations in Article 13 are also detached from workers in the labour context since they are limited to enabling users (so employers) to 'interpret the system's output and use it appropriately' (Kelly-Lyth, 2021).
Looking for further information?
Find a list of the research the iManage team conducted and drew on to develop the employee toolkit.
Algorithmic management:
The following article by Aislinn Kelly-Lynth and Anna Thomas and provides an overview of the impacts of algorithmic management. This Wired article also provides a shorter, lay summary of algorithmic management and the work of the iManage team.
Data protection law:
Between 2022 and 2024, the iManage team drew together resources for Data Protection Day curated around workers as data subjects:
- The Worker as a Data Subject: Resources for Data Protection Day (2022)
- Workplace Data Processing: Resources for Data Protection Day (2023)
- Data Processing at Work: Resources for Data Protection Day (2024)
In the following article, Regulating algorithmic employment decisions through data protection law, Halefom Abraha examines the extent to which the GDPR offers the necessary tools to protect workers from harm stemming from algorithmic management. In another piece, Halefom Abraha digs into Article 88 and EU Member State Employee Data Protection rules following the CJEU's decision in Hauptpersonalrat der Lehrerinnen concerning the live streaming of video classes in Germany during COVID. An additional piece by Halefom Abraha also focuses on Article 88 in upholding privacy in the workplace which is described as a 'doubled-edged sword'.
Discrimination law:
For a detailed account of the application of UK direct discrimination law to algorithmic decision-making, Directly Discriminatory Algorithms by Jeremias Adams-Prassl, Reuben Binns, and Aislinn Kelly-Lyth examines the case law on direct discrimination in the UK (with brief comparisons to the EU and US), showing its utility in the context of biased algorithmic decision-making. For an detailed account specific to the employment context, see Aislinn Kelly-Lyth's Algorithmic discrimination at work which applies direct and indirect discrimination law to algorithms at work, while advocating for the need for improved transparency for these existing protections to have any weight.
Platform Workers Directive:
For an analysis of the Platform Workers Directive with respect to algorithmic management, see Fortifying the algorithmic management provisions in the proposed Platform Work Directive by Michael Veale, Michael 'Six' Silberman, and Reuben Binns. A summary of the Platform Workers Directive can be found on blogs here and here, alongside blog posts (here and here and here) which criticises the limited scope of the legislation. Note, this research was conducted while the Platform Workers Directive was being debated, so its analysis is focused on the proposal/draft version of the Directive.
EU AI Act:
For an overview of the applicability of the EU AI Act to algorithmic management, see Aislinn Kelly-Lyth's article in the Comparative Labor Law & Policy Journal.
This site is part of a project that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 947806).
If you have any questions or feedback about this site or our research, please get in touch by emailing ai.work@law.ox.ac.uk.