Systemization of Knowledge (SoK): Human-Centered AI Safety for Youth

Most AI safety protections for youth remain unevaluated in real-world settings: The review of 100 empirical HCI studies found that while many countermeasures have been proposed to address AI risks, "few countermeasures have been implemented, and fewer still evaluated," meaning we lack evidence that these protections actually work when young people use AI systems in practice.

Current evaluations focus on technical performance rather than actual harm prevention: The paper identifies a critical gap where "existing evaluations often measure technical performance rather than protection from harm," suggesting that studies may show an AI system works correctly without demonstrating it actually keeps youth safe from identified risks.

Significant gaps exist in which AI risks receive attention and protection: The risk-countermeasure mapping reveals uneven coverage—some identified risks lack any proposed safeguards, while others have proposed solutions but no implementation or testing, leaving certain youth vulnerabilities largely unaddressed across schools, homes, and public services.

The field needs stronger real-world evidence before deploying AI systems to youth: With "most risks matched only with proposed/ideated countermeasures," the paper suggests the HCI and AI communities should prioritize implementing and rigorously evaluating protections in actual settings where children and youth use AI, rather than relying on theoretical solutions.