After a settlement between Meta and the Attorney-General of the United States of America,29 the agreement included a rare arrangement in addition to the payment of up to $18 billion and increased safety measures for children. According to the media, the Länder agreed not to maintain and use data on Meta in the training and testing of the child age identification model and to continue to prosecute under existing child privacy laws.
Settlement clause with limited immunity
Under the agreement, Meta is required to develop, train and start testing a model for the identification of users under 13 within one year of the entry into force of the agreement. The document does not directly state that AI must be used, but Meta's existing age recognition tool itself relies on AI technology.
The agreement also states that this exemption is limited to the training and testing of age identification models. Meta must not use data from users under 13 years of age for advertising orientation, marketing or algorithm optimization.
COPPA Law enforcement space is of concern
The Child Online Privacy Protection Act of the United States usually requires websites and applications to restrict the collection and preservation of personal information about children. According to the media, the settlement agreement not only sought to leave room for Meta's model training, but also included a stronger exoneration that the Attorney-General of the relevant state “completely, eventually and permanently” waived past, present or future COPPA and similar state law claims for the use of such data.
According to legal sources, Meta ' s pursuit of such protection is not entirely unusual, as businesses may indeed need to retain a small amount of data for compliance verification when dealing with the removal of requests, identification, etc. The problem, however, is that COPPA is a federal law and the main law enforcement agency is the United States Federal Trade Commission (FTC), which is not a signatory to the settlement agreement.
Data segregation and monitoring are more complex
The real challenge, according to external sources, is at the implementation level. Meta needs to isolate child behavioural signals and related data from other corporate systems and limit their use to the identification and removal of accounts for children under 13 years of age. However, data and models within large platforms are often interconnected and it is not easy to achieve complete technical and organizational isolation over the long term.
The agreement indicated that the independent auditor would be involved in monitoring Meta ' s compliance with the settlement clause. However, it remains unclear to the outside world what data Meta would retain for the training model, how much behavioural information it would involve, and how long it would be kept.
Several lawyers have indicated that the State Attorney can still be held accountable if Meta uses these data beyond the scope of the agreement. However, once disputes arise in the future, the focus of the case is likely to shift to whether the use of relevant data remains within the scope of the settlement agreement. This also means that subsequent enforcement and evidence may become more complex.
Dispute extended to AI data use issues
The controversy also touched on the broader issue of AI: models often require more user data support if they are to improve their ability to identify and judge. For Meta, identifying which accounts belong to children may also require more in-depth analysis of the use of social platforms by underage users.
According to external sources, this has further heightened the tension between child protection and model training. The settlement agreement attempts to draw an enforceable line between the two, but whether this line can be clearly monitored in the future remains the focus of outside concern.