From Age Gates to Accountability in AI Design

As artificial intelligence systems become increasingly enmeshed in everyday life, policymakers have intensified efforts to shield children from potential digital harms. Much of this regulatory attention thus far has focused on limiting children’s access to particular online environments, as reflected in recent international initiatives to restrict youth access to social media through age-based restrictions. While these measures are framed as child-protection efforts, they are not yet directed at artificial intelligence and instead prioritize questions of access over scrutiny of how digital systems are designed and deployed. This emphasis risks obscuring a more fundamental issue: whether digital systems, including AI, may inadvertently harm children even in the absence of malicious intent.

One potential framework for addressing this question can be found in the legal doctrine of disparate impact. Traditionally applied in anti-discrimination law, disparate impact analysis addresses practices that are neutral on their face but produce unjustified, disproportionate harm to protected groups. In recent years, scholars and policymakers have explored how this doctrine might apply to algorithmic discrimination based on race, gender, disability, or socioeconomic status. Far less attention has been paid to whether a similar analytical lens could be applied to minors as a distinct and structurally vulnerable group in the context of AI governance.

Disparate impact doctrine is premised on the recognition that harm can arise not only from intentional discrimination but also from systems and policies that fail to account for existing vulnerabilities. Under this framework, a practice may be deemed unlawful if it disproportionately affects a protected group and cannot be justified as necessary to achieve a legitimate objective, or if the same objective could be achieved through less harmful means. The doctrine shifts regulatory attention away from intent and toward outcomes, justification, and design alternatives.

This shift is particularly salient for children’s interactions with AI systems. Children are not merely younger versions of adult users. Their cognitive, emotional, and social development differs in ways that materially shape how they experience and are affected by technology. Legal systems have long recognized this reality in contexts such as consumer protection, advertising, education, and product safety. However, many AI systems continue to be designed primarily for adult users, relying on assumptions about autonomy, critical reasoning, and emotional resilience that do not hold for younger users.

As a result, features that may appear benign or beneficial for adults can have markedly different effects on children. Engagement-optimized recommendation systems can exacerbate attention fragmentation and compulsive use among minors. Conversational agents designed to simulate empathy and emotional availability can encourage dependency or displace human relationships. Personalized persuasive techniques can blur the line between assistance and influence in ways that children are less equipped to recognize or resist. These effects are often cumulative, subtle, and difficult to trace to individual instances of harm, which makes them poorly suited to existing regulatory frameworks that focus on discrete content violations.

Applying a disparate impact lens would shift responsibility in AI development away from intent and toward outcomes, asking whether foreseeable and disproportionate harms to children are justified as necessary to achieve objectives such as engagement or usability, and requiring developers to show that less harmful design alternatives are not reasonably available. This approach addresses key limits of access-based regulation: bans and age restrictions are difficult to enforce, easy to circumvent, and ignore evidence that restricting specific platforms often just shifts children’s screen time elsewhere rather than reducing it. A disparate impact framework instead targets the structural features of AI systems and how their effects are distributed across different user populations.

Elements of this design-oriented approach are already emerging in various jurisdictions. The European Union’s Digital Services Act and AI Act incorporate concepts of systemic risk and heightened protections for vulnerable users. The United Kingdom’s Age Appropriate Design Code embeds the best interests of the child into product design expectations. Australia’s treatment of certain AI companions as high-risk technologies reflects growing recognition that some systems pose unique concerns for minors. However, these initiatives often lack a unifying legal rationale that clearly articulates why children warrant distinct protection beyond content moderation or access control.

As governments continue to debate how best to protect children in digital environments, the focus should not rest solely on whether minors are permitted to access particular technologies. It should also encompass whether those technologies are built in ways that unfairly burden young users. Disparate impact analysis offers a framework for asking that question systematically and for aligning responsible AI development with the realities of children’s lived experience. Crucially, such measures would shift accountability upstream, encouraging developers to address risks during the design and deployment phases rather than after harms have already materialized.

In an era in which AI systems increasingly shape how young people learn, communicate, and relate to the world, governing these technologies requires more than access restrictions. It requires a clear-eyed assessment of how design choices distribute risk and responsibility. Extending disparate impact principles to minors may serve as a step toward meeting that challenge.