The problem of AI use in social work with children and families

Alongside other welfare services, advanced digital technologies, including Artificial Intelligence (AI), continue to be rapidly assimilated into welfare sectors including statutory social work. Indeed, social work with children and families is increasingly becoming a significant site of experimentation for digital technologies. Driven primarily by a combination of technological innovations and managerial imperatives for efficiency and accountability, the introduction of AI into social work practice represents a paradigm shift in how professionals engage with children and their families. AI-enhanced technologies now influence how front-line practitioners communicate, process sensitive information, and fulfil core tasks such as assessing and evaluating complex needs and the planning of interventions (Haider et al, 2025; Rothera and Macdonald, 2025; Sumskiene et al., 2026).

GenAI, which has the capacity to learn independently, is increasingly being used by social workers to fulfil key roles in the United Kingdom. AI-enhanced technologies such as predictive analytics, generative language models, digital scribes and algorithmic decision-making are also championed as reliable solutions to organisational inefficiencies, finite resources and the need to improve core services (McQuillen, 2025; Pelletier et al., 2025). However, the rapid expansion of digitalisation agendas and AI also has the potential to extend the remit of the state further into the lives of minoritised families. This includes, not least, historic assumptions that the children of working-class poor and Black families represent the property of the state (Herbert, 2026).

Nevertheless, concerns about AI-enhanced technologies persist. Although these tools may appear convenient and potentially empowering, their integration into professional discursive domains risks over-reliance, weakens meaningful ethical practice, and may reduce the rights of both ‘service users’ and social workers. Boothe (2026), for example, reports that AI tools have produced harmful errors in social work records, including false alerts and instances of “gibberish”. Despite being promoted as objective, AI systems have also been shown to reproduce forms of prejudice, including those associated with ‘race’, social class and gender (Avraamidou, 2024; Sumskiene et al., 2026).

This article focuses on two interrelated concerns. First, it examines the difficulties involved in governing AI-enhanced technologies ethically within social work with children and young people. Second, it considers how AI-enhanced systems may intensify further long-established patterns of prejudice in work with minoritised families and their children.

Ethical issues and AI use within social work

Social work with children and families routinely involves ethical tensions, particularly between maintaining the rights of parents or carers and protecting children deemed to be at risk. Despite adherence to Kantian-inspired professional ethical codes, there is limited evidence that social workers consciously or consistently use such codified principles in practice. Alongside limited time and resources, this is partly because translating ‘abstract ethical principles into concrete decisions and actions’ remains difficult within a complex, highly stressful and ‘messy’ moral role (Congress, 2010: 25). Despite this, far from representing benign rhetorical constructs, professional codes and other interpretations of normative ethics, can still maintain a strong influence at street-level. For example, core ethical principles such as autonomy and duty can be utilised to fortify neo-liberal policy objectives through market-friendly concepts such as rational choice and autonomy. They can also enhance professional and managerial powers (including via core legislation), not least over parents and young people compelled to use social services as well as staff (Estes et al, 2003; Petrie, 2009).     

AI and the continued encroachment of digital technologies within welfare retain the potential to significantly extend the regulation and surveillance of social work’s core client groups. In principle at least ethics offers a means by which AI can be contained including by offering an enlightened sense of moral purpose in work with families. However, the evidence thus far – including in key associate fields of welfare such as health care – suggests that numerous logistical, technical and business-orientated obstacles remain. Arbelaez Ossa et al (2024a), for example, examined guidelines as part of a discourse analysis of the widespread use of AI within health and social care sectors. By focusing on inter-governmental, governmental and professional organisations, the researchers’ highlight the prevalence of overt hype and enthusiasm generated about the capacity of AI to enhance the quality of care and patient wellbeing. Moreover, intense pressure was placed upon service providers and professionals from business and other provider representatives, including to instil AI-enhanced technologies of care within hospitals, clinics and everyday practice. Invariably this affirmative discourse understated any possible risks which AI posed and a tendency to overlook and exclude patient perspectives was also highlighted. The authors highlight how such commercial drivers invariably risk pushing principles which are detached from altruistic care values, to instead privilege pervasive technological and business orientated doctrines that ‘fail to align with the needs of society’ (Arbelaez Ossa et al, 2024a: 18).

A recent national survey (including 155 front line social workers) and comprehensive literature review undertaken for Research in Practice and Social Work England by Rothera and Macdonald (2025) details the increasingly common place use of GenAI by social workers. This includes as virtual assistants or chatbots and for transcription or case recording. Whilst many practitioners believed that AI could reduce bureaucracy and potentially increase time to spend with children and their parents, several concerns were raised by participants and the authors of the study. Among others, these included a training gap about how to use AI as part of the social work role, with no formal preparation commonplace for staff. Other concerns related to privacy issues: 

What AI does with information, and how it processes it and stores it, is often unclear to the user. This is particularly concerning in relation to AI applications originating from outside the UK and not subject to GDPR or data protection laws. (ibid: 20)

By drawing on interviews with 39 social workers, Bruff and Groves (2026) raise ethical concerns about the rapid adoption of AI transcription tools within local authorities including work with children and families. Although many practitioners often valued any time-saving benefits offered, views on their reliability tended to vary. The researchers also criticised the limited, ‘light touch’ approach to ethics and any evaluations of these tools and noted that local authorities tended to prioritise efficiency gains over the impact of AI technology on people using services. Wilkins and Benett (2026) examined the use of ChatGPT in child safeguarding through 12 anonymised case vignettes, comparing assessments produced by 581 children and family social workers with those generated by an AI model to evaluate each group’s ‘judgemental accuracy’. Although the AI-enhanced model was faster and demonstrated some comprehension, its wider use raised significant ethical concerns, particularly regarding its limited ability to grasp case complexity, contextual reasoning and practice sensitivity. The authors also stress the ongoing challenge of managing access to sensitive data, as well as the continued importance of communication and interpersonal skills in casework—areas in which AI is likely to struggle.

Wider concerns have also been raised about the practical difficulties of embedding ethics in high-pressure settings where speed and efficiency are prioritised, and where ethics can commonly be viewed as an organisational burden that ‘slows things down’ (Arbelaez Ossa et al, 2024b). More broadly, because ethics is itself discursive and contains often unresolved questions including those associated with social work, it cannot therefore provide the kind of unequivocal algorithmic guidance that AI systems require. Expecting AI to process unresolved questions about poverty, neglect or discrimination and then apply these within complex social fields of practice therefore appears risky, if not futile.

Prejudice and datafication

AI surveillance technologies are increasingly being integrated into social work practice to assess risk, monitor behaviour, and decision-making processes. One of the most widely used applications is predictive analytics, which employs machine learning algorithms to identify individuals at risk of adverse outcomes such as child abuse, homelessness, or criminal behaviour (Munroe, 2019). A good example is the Allegheny Family Screening Tool (AFST) in Pennsylvania which uses historical data to predict the likelihood of child maltreatment and guide caseworkers in making intervention decisions (Chouldechova et al., 2018). Similarly, AI-driven biometric tracking and facial recognition systems are now used in welfare programs to prevent fraud and monitor service users’ behaviour (Eubanks, 2018).

The expansion of increasingly sophisticated technology raises concerns about ever more prejudice faced by already structurally disadvantaged minority groups. Attention has previously been drawn to the ways in which English child protection systems are built upon historical ties to colonialism, leading to the meticulous regulation of the parenting capacity of minority ethnic mothers. Indeed, Herbert (2026: 4-5) has argued that the children of black mothers and who live in poverty within former Imperial powers such as Britain represent the state’s property, with their mother’s morally scrutinised regarding their parenting capacities and any children deemed as needing to ‘be controlled to become an ideal citizen’. Alongside enduring poverty this doctrine generated through the state apparatus has led to a significant over representation of black children in care – as well as an associated and indeed often normalised assumption within wider welfare – that black mothers represent a risk to their children and are ‘alien, different and inferior’.

Far from supporting objective and fair assessments or purposeful engagements with state representatives, AI systems offer a powerful technical means by which to extend existing stereotypes and forms of prejudice. For example, current AI systems struggle to accommodate diversity leading to algorithmic bias. This results from AI systems focusing on historical data sets which support societal biases to a point where they replicate and potentially harden structural forms of inequality and exclusion. These dynamics can quickly lead to the acute surveillance of minoritised groups such as black adolescent working class males disproportionately targeted for intervention (Mohamed et al, 2020; Redden et al., 2020). According to Noble (2018), for example, large language models, which focus their attention on substantial datasets and excel at coding, summarisation or translation, risk further increasing forms of racism. Rothera and Macdonald’s (2025) recent survey which included responses from 155 social workers highlighted numerous examples of concerns raised by practitioners about the continued reliance upon AI enhanced technology. One issue highlighted by some of the practitioners related to bias and the risks posed by the possible further exclusion of already marginalised communities. Indeed, as one practitioner noted:

“We are going into this too quickly and not thinking enough about the inbuilt biases of AI, e.g., marginalised communities may become more marginalised. Social work is a nuances activity that involves all your body and senses and not just the fingers you use to type.”

Relatedly, Mohamed et al (2020) have again highlighted the limited capacity for ethical foresight on behalf of AI systems – which leads to what the authors’ term ‘algorithmic colonialism’ – in which advanced technological systems lack capacity to be able to meaningfully understand or contextualise unseen yet deeply felt dynamics faced by structurally excluded and oppressed groups. Briefly there are three specific ways by which ‘algorithmic colonialism’ is manifest:

1. Lack of Ethical Foresight: AI often ignores the historical trauma of marginalized groups.
2. Asymmetries of power: The concentration of AI power in a few companies mirrors colonial monopolies.
3. Epistemic Violence: AI models prioritize Western “rationality” and English-language data, effectively erasing other forms of knowledge.

Another trend which AI-enhanced technology inevitably risks amplifying further is that of the datafication of social work’s core client groups, including children and their parents. This ongoing trend can lead to young people quickly being reduced to data points – which adds to past criticism that the subjects of a risk-orientated, fragmented and under resourced child protection industry – remain ever more objectified within bureaucratic systems that overlook the challenges and complexity of childhood. Ongoing criticism, for example, has highlighted how children are increasingly invisible within risk-averse and process-driven formal care systems (for example, Petrie, 2009; Ferguson, 2017; Morris et al, 2018). Furthermore, such dehumanising processes are strengthened by AI surveillance and technologies of care which prioritise risk scores above personal perspectives whilst reducing any capacity for expressions of voice and self-advocacy among people compelled to use state sponsored welfare services (Zuboff, 2019). Additional concerns have also been raised about surveillance creep, including how any data collected and processed by social workers can be shared with other organisations, including the police and government agencies. This outcome is likely to encourage further scepticism felt about social services by minoritised groups, especially children, young adults and families more prone to negative interventions (Redden et al., 2020).

Conclusions

It appears highly likely that AI-enhanced technologies will continue to expand within social work. The danger is that already established patterns of regulation and control for structurally disadvantaged children, and their families may further intensify. Alongside the datafication of children or parents, as well as structurally or historically induced forms of prejudice, there remains a likelihood that professional powers over young people will increase. AI must not therefore be viewed as a neutral technical innovation. Rather, it represents a significant development in the ongoing technologization of welfare and social work practice (McQuillen, 2022; Bruff and Groves, 2026).

A key challenge is not whether AI should be used, but how its use can be governed in ways that remain consistent with progressive social work values. In practice, social work with children and families remains relational, and effective practice depends upon traits including trust, empathy, communication and contextual understanding. These capacities cannot be automated and at present remain beyond the reach of even the most sophisticated AI systems. Ethical judgements also cannot be outsourced. While technology may assist practitioners in organising information and identifying potential concerns, decisions affecting children and families require human responsibility, accountability and moral reasoning. AI systems are also clearly not free from prejudice, with historical data frequently containing embedded inequalities and discriminatory patterns. Without scrutiny, algorithms risk reproducing and intensifying these injustices. Finally, the expansion of AI raises broader questions about power and governance. Technologies introduced in the name of efficiency may simultaneously strengthen surveillance and managerial control, and such developments require ongoing democratic scrutiny. Social workers of the future will need not only digital skills but also the capacity to critically evaluate technological systems and recognise their limitations. Professional education and training must prepare practitioners for this new landscape.