The National Association for Youth Justice (‘NAYJ’) published a briefing paper in September 2026, titled Predictive Analytics and Algorithmic Technologies in Youth Justice: Innovation at what cost?.
The briefing comes after the Ministry of Justice’s 2025 AI Action Plan for Justice, drawing attention to key concerns and calls for action.
Details
Key Concerns
- Labelling and stigma. Predictive systems often identify children ‘at risk’ of future offending by applying labels such as ‘deviant’, ‘high risk’, or via algorithmic scores with similar implications. These often only further the marginalisation and secondary deviance of young people (by promoting that very narrative).
- Desistance. Predictive systems focus on – and are biased towards – risk factors, without considering other (unmeasurable) components that further patterns of desistance in children’s lives, such as supportive adults or positive impact activities.
- Transparency and accountability. The internal decision-making processes of these tools remain behind technological barriers, possibly immune to proper scrutiny and consequent legal challenge. In turn, decision-makers trust and rely too much on algorithmic outputs, without sufficient scepticism (known as automation bias), thus exacerbating the problem. Additionally, these obscure processes risk undermining the security of children’s data.
Call for action
- Increased transparency regarding the operation, design, and purpose of predictive tools – supported by independent studies, monitoring efforts, audits, and publications to allow scrutiny.
- Providing child-specific impact assessments which focus on these tools’ compliance with human and children rights standards and help ensure fair decision-making processes by professionals using these tools.
- Social inclusion and participation of children in the design, implementation and evaluation of these tools.
Commentary
- Using predictive tools in youth justice presents new risks and challenges for the profession, which practitioners should keep in mind. These include:
- the harmful effects of overreliance on predictive tools, including automation bias, labelling and stigma issues;
- what factors such tools might fail to consider;
- whether such tools could have disproportionate effects across protected characteristics (e.g., race, gender or disability), and
- whether the decision-making process of these tools can be adequately scrutinised.
For related information, visit NAYJ.