1. Introduction and problem

In early childhood and basic education schools a seductive promise circulates: artificial intelligence “personalizes” learning, “frees” the teacher, and “makes visible” what the human eye cannot reach. That promise usually silences its condition of possibility. To personalize, a datum is required. To free, a metric. To make visible, a camera, a microphone, or a continuous behavior log. The datum is not abstract. It is the voice of a four-year-old girl, the face of an eight-year-old boy, the point a teacher assigns because a child “worked hard” or “spoke out of turn.” Those who produce that datum cannot read a privacy policy, cannot negotiate with a vendor, and cannot opt out of being in the classroom.

The problem is not hostility to technology. It is the confusion among three objects. The first is an empirical finding: what is observed when a platform quantifies conduct, when children’s voice is recorded, or when facial recognition is trialed on school grounds. The second is a normative framework: what the Convention on the Rights of the Child in the digital environment, the GDPR, COPPA, UNICEF, UNESCO, and data-protection authorities require. The third is a pedagogical inference: what should not be delegated in the classroom even when marketing says “parental consent” or “educational benefit.” Mixing them produces a pedagogy of capture: the school “uses AI” without being able to say what is extracted, who retains it, for how long, and with what right to say no.

This article’s thesis is restrictive. Educational AI in the early childhood and basic education classroom does not personalize without cost; it extracts voice, image, and behavior from subjects who cannot consent. That threshold—the non-delegable—is not resolved by a box signed by an adult or by the fiction that the school always acts as the parent’s agent. Lupton and Williamson (2017) named the “datafied child” and warned of the scarcity of specific instruments against dataveillance. Since then, the classroom has concentrated behavior platforms, learning analytics, voice recording for automatic recognition and, in some districts, facial biometrics. This article does not treat parental mediation at home (19 August extra axis), nor curricular literacy for students (20 August night), nor disability inclusion (20 August, 5:00), nor UNESCO teacher competence (19 August morning), nor play or PopBots (18 August), nor adaptive tutors (16 August). The object is the classroom: the space of compulsory presence where capture is naturalized.

Where the source does not measure learning improvement, this article does not affirm it. Where it does not measure that data protection “is already solved,” it does not promise it. The gap organizing the work is consent under asymmetry: the child cannot refuse school; the parent often cannot refuse the platform the school chose; the teacher is trapped between care and quantification (Lu et al., 2021).

2. State of the art: datafication, privacy, and the classroom as enclosure

Three strata should be separated. The first is conceptual: what it means to datafy childhood at school. The second is normative: which frameworks set privacy, consent, and limits on biometrics or AI. The third is empirical: what has been documented in classrooms and school platforms, not with the home as the axis.

In the conceptual stratum, Lupton and Williamson (2017) critique datafication and dataveillance from gestation through the school years and conclude that there is little evidence of specific instruments to safeguard rights against those practices. Their article is not a classroom trial; it is retained as a critical frame of the “datafied child.” Manolev, Sullivan, and Slee (2019) move that frame to ClassDojo and argue that the datafication of discipline intensifies and normalizes surveillance, produces a performative culture, and operates as behavior control. Williamson (2017), cited in that line, had linked ClassDojo to psycho-policy and persuasive social-emotional learning; here it is not taken as a nuclear case, but as antecedent of the debate.

In the normative stratum, the Committee on the Rights of the Child (2021), in General Comment No. 25, affirms that Convention rights apply in the digital environment. On privacy (paras. 67–73), it requires that any interference be lawful, legitimate, proportionate, with data minimization and the best interests of the child; that consent—by the child or the holder of parental responsibility according to age and evolving capacities—be informed and free; and that access, rectification, erasure, and objection be available. UNICEF (2021), in Policy guidance on AI for children 2.0, requires protecting data and privacy, adopting privacy by design, minimizing “invisible” processing, and not allowing childhood data to follow the person into adulthood. GDPR Recital 38 recognizes specific protection for children; Article 8 regulates a child’s consent regarding information society services (threshold of 16, reducible to no less than 13); Recital 43 warns that consent is not free under a clear imbalance, particularly when the controller is a public authority; Recital 71 states that automated decision-making based on profiling should not concern a child. Miao and Holmes (2023) warn that the absence of national regulation leaves user privacy unprotected and propose mandating data-privacy protection and an age limit for independent conversations with generative AI. UNESCO’s Recommendation (2021) elevates privacy and data protection to a principle. Regulation (EU) 2024/1689 (Annex III, point 3) classifies as high-risk AI systems intended to determine access or admission to educational institutions, evaluate learning outcomes, assess the appropriate level of education, or monitor and detect prohibited student behavior during tests. The FTC (2022) recalls that, in the school context, COPPA-covered edtech operators may not use student information for commercial purposes unrelated to the requested educational service. The American Academy of Pediatrics (Radesky et al., 2020; AAP, 2026) insists that children have difficulty understanding data collection and cannot meaningfully consent, and calls for privacy by default and avoiding surveillance and targeted advertising. The ICO (2023) clarifies that schools are not “information society services” under the Children’s Code, but that edtech providers may be when they determine their own purposes (research, marketing, product development).

In the empirical classroom stratum, qualitative evidence on ClassDojo (Lu et al., 2021; DiGiacomo et al., 2021; Manolev et al., 2024) documents behavior quantification, care/control tensions, and metric power, without measuring academic learning as a platform effect. Voice corpora (Pradhan et al., 2024) document scale and institutional consent procedures, not pedagogical improvement. The Skellefteå decision (Integritetsskyddsmyndigheten, 2019), the CNIL position (2019), and New York’s determination (New York State Education Department, 2023), together with Galligan et al. (2020), document disproportion, invalid biometric consent, or prohibition of school facial recognition. The gap is not the absence of norms: it is the school naturalization of capture under the lexicon of personalization.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate an effect size of “AI and learning,” nonexistent in a homogeneous way across a teacher-interview study, a speech corpus, and an administrative sanction, but to articulate the non-delegable threshold with verified sources. Inclusion criteria were: (a) focus on classroom, school, or platform used on school grounds, not parental mediation at home as the axis; (b) preferred publication 2021–2026, with justified exception for standing frameworks (Lupton & Williamson, 2017; Manolev et al., 2019; McReynolds et al., 2017; GDPR, 2016; COPPA; UNESCO, 2021; UNICEF, 2021; Committee on the Rights of the Child, 2021; Galligan et al., 2020; Integritetsskyddsmyndigheten, 2019; CNIL, 2019); (c) document type peer-reviewed journal or proceedings with DOI, archived university report, official decision or release of a data-protection or education authority, or legal text on EUR-Lex / FTC / NYSED / ICO; (d) access to DOI, publisher page, or official URL confirming authors, year, title, and scope. Excluded as nuclear cases were parental mediation and home toys (except brief saturation), student curricular literacy, disability inclusion, UNESCO teacher competence, PopBots/genAI play, and adaptive tutors/assessment.

The search was executed on 21 August 2026 (around 01:06, America/Mexico_City) on DOI pages, ACM Digital Library, Taylor & Francis, Springer, ScienceDirect, EUR-Lex, FTC, EDPB/IMY, CNIL, NYSED, UNESDOC, UNICEF Innocenti, ICO, Deep Blue (University of Michigan), and ACL Anthology. Each cited source was verified against at least one of those pages. The corpus was organized into three nuclear cases: behavioral datafication (ClassDojo); children’s voice in educational contexts; facial biometrics in schools. McReynolds et al. (2017) is used only as saturation of recording opacity, not as a nuclear case.

Analysis distinguished three enunciative statuses. Empirical finding: what was observed or decided in the sample or file. Normative framework: what an organization prescribes. Pedagogical inference: the translation to early childhood and basic classrooms, marked as such. Limits are those of any narrative review (section 9).

4. Case 1. ClassDojo and the datafication of classroom behavior

ClassDojo is a communication and behavior-management platform that lets teachers quantify, record, and communicate points linked to desired or undesired behaviors, alongside portfolios and family messaging. Manolev, Sullivan, and Slee (2019) critically examined how the system datafies discipline and student behavior. They argue that it intensifies and normalizes surveillance, creates a culture of performativity, and serves as a mechanism of control. The status is critical analysis of design and disciplinary rationalities, not a controlled learning trial. They do not measure whether points improve reading or mathematics; they show, conceptually, how conduct becomes number, comparable and governable.

Lu, Dillahunt, Marcu, and Ackerman (2021) published in Proceedings of the ACM on Human-Computer Interaction (CSCW) an interview study with twenty K-8 teachers using ClassDojo. Empirical finding: teachers deploy the tool both to control and to care; at the same time they are exposed to the gaze of parents, administrators, and students, and perform “data work”—recontextualizing bodies and needs reduced to points, and sometimes manipulating points to resist surveillance expectations. The article does not claim ClassDojo improves learning; it documents the tension of (in)visibility and teacher autonomy under datafication.

Lu, Marcu, Ackerman, and Dillahunt (2021), in DIS, analyze the same qualitative line (twenty K-8 teachers) under the angle of coding bias. Empirical finding: ClassDojo use risks measuring, codifying, and simplifying psycho-social factors that drive behavior, operating as a “Band-Aid” for deeper issues and potentially perpetuating inequality. They propose three design considerations: provide context, expose bias, and challenge what counts as “normal.” DiGiacomo, Greenhalgh, and Barriage (2021), in TechTrends, offer a mixed-methods study in a southeastern U.S. state on how students and principals understand ClassDojo. Empirical finding: there is a perception of mediation of student–teacher and student–parent relationships; privacy concerns appear in media and literature but remain under-studied in everyday experience. Manolev, Sullivan, and Tippett (2024), in the British Journal of Sociology of Education, examine teachers’ practices with ClassDojo through critical qualitative inquiry. Empirical finding: practices operate via control techniques centered on metrics; the authors invoke “metric power” and argue that ClassDojo metrics act as narrowing pedagogical devices that fix attention on measurement and reshape school discipline under neoliberal governing rationalities. They also do not measure academic achievement.

Evidence status. Accumulated empirical finding: in K-8 classrooms, ClassDojo quantifies conduct, shifts psycho-social complexity toward points, exposes teachers and students to cross-surveillance, and reconfigures discipline as government by metrics. It is not a finding of learning improvement. It is not a finding that privacy “is covered” because an online policy exists: DiGiacomo et al. (2021) and critical literature precisely note opacity and under-research. Pedagogical inference, marked as such: in early childhood and the first cycle, turning the school day into a points board is already datafication of compulsory presence. Even without a camera or microphone, there is a behavioral profile. That profile is not “pedagogical personalization” while—and here it is not shown—a causal link to learning remains undemonstrated; it is capture of the conduct of someone who cannot refuse the classroom.

5. Case 2. Children’s voices: corpora, recording, and the microphone threshold

A child’s voice is potentially biometric data and, under COPPA, an audio file containing a child’s voice is personal information (Federal Trade Commission, n.d., FAQ F). In educational contexts that voice is collected for virtual tutors, automatic recognition of children’s speech, classroom language analytics, or research. Pradhan, Cole, and Ward (2024) describe the My Science Tutor (MyST) corpus: approximately 400 hours of conversational speech from about 1,300 third-to-fifth-grade students in sessions with a virtual science tutor (about 10,500 sessions and 230,000 utterances; the release includes over 100,000 transcribed utterances). The project reports IRB review, parental consent and student assent, and distribution of anonymous data. Empirical finding: a school-scale children’s speech corpus aligned to science instruction exists; the authors report word-error rates of models trained on the corpus and offer it for speech-recognition research. It is not a finding that the virtual tutor improves science learning in this corpus-description paper: the object is the data resource and ASR, not a pedagogical efficacy trial.

Dutta, Irvin, and Hansen (2025) explore, in the International Journal of Human-Computer Studies, discrete speech units as anonymous encoding for automatic recognition aimed at school-aged and preschool children. Empirical finding: a discrete ASR model trained on MyST reaches a 15.7% word error rate and performance comparable to much larger end-to-end models; the authors argue that the sequence of discrete units makes reconstructing the original waveform virtually impossible, introducing a degree of privacy. They also report promising results recognizing WH-words, nouns, verbs, and pronouns in a case study of teacher–child interactions in a childcare facility. Status: technical finding of privacy by design in child ASR; it is not a finding that the classroom “is already protected” if raw audio is recorded, nor of curricular improvement.

As saturation—not as a nuclear case—McReynolds, Hubbard, Lau, Saraf, Cakmak, and Roesner (2017) showed, in interviews with parent–child pairs and connected toys, that children often did not know the toy was recording or that others might hear what was said. It is cited only to underline the phenomenological opacity of recording: if that occurs with a visible toy at home, the pedagogical inference—marked as such—is that a vest or classroom microphone, presented as “to research language” or “to personalize,” may be even less intelligible to an early childhood learner. That saturation does not shift the axis to the home.

Pedagogical inference: recording children’s voices in the classroom crosses a threshold distinct from noting a grade. Voice identifies, enables affective inferences, and can train commercial or research models beyond the stated purpose. UNICEF (2021) requires minimization and short retention. The Committee on the Rights of the Child (2021) requires defined purpose and no unnecessary retention. Dutta et al. (2025) show technical minimization paths; they do not show that school platforms adopt them by default. Where the microphone is continuous and the child cannot switch it off, parental consent—or school consent as agent under COPPA guidance—does not erase the imbalance Recital 43 of the GDPR flags toward public authorities.

6. Case 3. Face and biometrics: when the classroom becomes a sensor

Facial recognition in schools has produced a corpus of administrative decisions, policy reports and, to a lesser extent, empirical deployment studies. Integritetsskyddsmyndigheten (2019)—then Datainspektionen—fined the municipality of Skellefteå SEK 200,000 for a three-week pilot with 22 upper-secondary students using facial recognition to monitor attendance. Authority finding: sensitive biometric data were processed without a valid exception; consent was not an adequate basis because of the power imbalance between school and students; processing was more intrusive than necessary (GDPR Art. 5); an adequate impact assessment and prior consultation were missing (Arts. 35 and 36). It was Sweden’s first GDPR fine. It is not a learning trial: it is a legality file.

The CNIL (2019) ruled on the experimentation of facial-recognition “virtual gates” in two lycées (Nice and Marseille). It considered the scheme contrary to GDPR proportionality and minimization: concerning mostly minor pupils, and with less intrusive alternatives available (e.g., badge control), facial recognition to ease and secure access appeared disproportionate and unlawful. Galligan, Rosenfeld, Kleinman, and Parthasarathy (2020), in the University of Michigan report Cameras in the Classroom, analyze implications of school facial recognition through analogical case comparison. They anticipate exacerbation of racism, normalization of surveillance and erosion of privacy, narrowing of the “acceptable” student, commodification of data, and institutionalization of inaccuracy; they recommend banning use in schools. New York State Education Department (2023), after the Office of Information Technology Services report under State Technology Law § 106-b, determined to prohibit purchase or use of facial recognition technology in the state’s schools; other biometrics remain local decisions conditioned on privacy, civil rights, effectiveness, and parental input.

Evidence status. Normative-administrative finding: in Sweden, France (CNIL position), and New York, school facial recognition has been sanctioned, declared disproportionate, or banned. Policy-report finding: Galligan et al. (2020) argue structural risks without claiming an RCT. It is not a finding that AI cameras improve safety or learning: those variables are not demonstrated in this corpus. Pedagogical inference: if facial biometrics fails the proportionality test for attendance or access in secondary education, it fails a fortiori in early childhood, where the face is developing identity, the child does not understand the processing, and the less intrusive alternative (roster, badge, teacher presence) is trivial. Regulation (EU) 2024/1689, by listing certain educational AI uses as high-risk, reinforces—as framework, not as evaluation of a concrete product—that automated surveillance and assessment in education are not “innocent innovation.”

7. Inferential framework: the threshold of the non-delegable

The framework that follows is this article’s pedagogical inference, anchored in the cases and verified instruments. It is not a new international standard. It distinguishes five admissibility tests. If a proposal for “AI in the classroom” of early childhood or basic education fails them, it is not deployed.

7.1. Compulsory presence test. The classroom is not an information-society service the child enters by choice. GDPR Recital 43 and the Skellefteå decision (Integritetsskyddsmyndigheten, 2019) show that consent weakens or is invalidated under power imbalance. Inference: presenting a behavior platform, continuous microphone, or biometric camera as an “option” is rhetoric; the real option is often to accept or be left out of ordinary school activity. The non-delegable begins here: freedom is not pretended where schooling is compulsory.

7.2. Purpose and minimization test. UNICEF (2021) and the Committee on the Rights of the Child (2021) require defined purpose, minimization, and short retention. The FTC (2022) limits edtech use to the contracted educational purpose, without marketing. Inference: if the vendor retains voice or behavior to train its own models, improve product, or profile, capture is no longer a “digital extension of the school” (ICO, 2023). In early childhood, default minimization is not to record; if recording for research, Dutta et al. (2025) illustrate techniques of waveform irreversibility—not that vague policies suffice.

7.3. Non-substitution of pedagogical judgment by metrics. Manolev et al. (2019, 2024) and Lu et al. (2021) document how the point replaces complexity. Inference: behavior or “engagement” analytics are not sold as learning personalization while evidence of learning is absent. This article did not find that evidence in the nuclear corpus. Affirming it would invent a result.

7.4. Biometrics and proportionality test. CNIL (2019), Integritetsskyddsmyndigheten (2019), Galligan et al. (2020), and New York State Education Department (2023) converge: the face at school demands a high threshold or a ban. Inference: in early childhood and basic education, facial recognition for attendance, access, or “attention” fails proportionality when alternatives exist. Regulation (EU) 2024/1689 reinforces the high-risk character of certain educational AI systems.

7.5. Child agency and the right to say no. The Committee on the Rights of the Child (2021) requires informed free consent, access, rectification, objection, and information in child-friendly language. AAP (2026) states that children cannot meaningfully consent to collection. Inference: if the child cannot say no without school sanction, and the parent cannot withdraw the datum without excluding the child from class, there is no consent; there is acquiescence. The non-delegable is the right not to be turned into a training stream.

Operationally, the framework admits: (a) local tools without commercial profiling; (b) aggregated erasable data; (c) research with IRB, verifiable consent, and demonstrable minimization techniques; (d) human evaluation of conduct without a permanent board. It rejects: (e) routine school facial biometrics; (f) continuous voice recording without narrow purpose and deletion; (g) behavior metrics as a proxy for learning; (h) the fiction that “the school already consented” closes the ethical debate.

8. Discussion

Three tensions organize the discussion. The first is between personalization and extraction. Education marketing speaks of adapting; the cases show quantifying conduct (Manolev et al., 2024; Lu et al., 2021), accumulating voice (Pradhan et al., 2024), and attempting to read faces (Integritetsskyddsmyndigheten, 2019). Without learning evidence in this corpus, “personalize” functions as a euphemism for capture. Miao and Holmes (2023) warn of unprotected privacy amid generative AI speed; that warning holds a fortiori for AI operating on children’s bodies in the classroom.

The second tension is between formal consent and free consent. COPPA and FTC guidance allow, within limits, the school to act regarding parental consent for non-commercial educational uses (Federal Trade Commission, 2022; COPPA FAQs). The GDPR, by contrast, relativizes consent before authorities and effectively bars basing sensitive biometrics on a school “yes” (Integritetsskyddsmyndigheten, 2019). Inference: complying with an FAQ does not exhaust General Comment No. 25 or the best interests of the child. The school can be an administrative agent and still fail the ethical test of the non-delegable.

The third tension is between privacy by design and privacy by press release. Dutta et al. (2025) show child ASR can be designed with waveform irreversibility. UNICEF (2021) requires privacy by design. Behavior platforms and many school cameras do not demonstrate, in the cited evidence, that design. A handbook notice is not minimization. Galligan et al. (2020) warn of surveillance normalization: each “small” pilot trains the institution to watch without being watched.

If the object is the early childhood and basic classroom, the teacher need not be a lawyer. The teacher needs the authority to say that a continuous microphone, a perpetual points board, or a biometric camera is not pedagogy while legitimate purpose, proportionality, and measured educational benefit remain undemonstrated. Replacing that judgment with the vendor’s license is institutional capture.

9. Limits

This review is narrative. It does not apply a full PRISMA protocol or estimate combined effects on learning. ClassDojo studies are qualitative or mixed-methods with limited samples (e.g., twenty teachers in Lu et al., 2021); they are not randomized achievement trials. MyST and Dutta et al. (2025) measure ASR and technical privacy, not curricular attainment. The Skellefteå, CNIL, and NYSED decisions are acts of authority, not ethnographies of early childhood classrooms. McReynolds et al. (2017) is home saturation, not school evidence. Geography is biased toward the United States, Sweden, France, the United Kingdom (ICO), and UN frameworks; no Latin American trial of biometrics or ClassDojo in early childhood meeting these criteria was located with the same verifiable openness. That absence is a gap, not proof of nonexistence. UNICEF, CRC, UNESCO, GDPR, COPPA, AAP, and Regulation (EU) 2024/1689 frameworks are prescriptive: their authority is normative. Section 7 inferences are ethical-pedagogical threshold hypotheses, not evidence of national implementation. This article does not claim that all analytics are unlawful; it claims that, in the verified corpus, neither learning improvement nor resolved privacy was demonstrated.

10. Conclusions

Artificial intelligence in the early childhood and basic education classroom does not personalize without cost. It extracts voice, image, and behavior from girls and boys situated in a regime of compulsory presence. Three families of verified evidence sustain the argument. ClassDojo shows datafication of discipline and metric power without proof of learning (Manolev et al., 2019, 2024; Lu et al., 2021; DiGiacomo et al., 2021). Children’s speech corpora and models show scale of capture and, at the same time, technical minimization paths that do not equal protection already achieved (Pradhan et al., 2024; Dutta et al., 2025). School facial recognition has been sanctioned, declared disproportionate, or banned in weighty decisions and reports (Integritetsskyddsmyndigheten, 2019; CNIL, 2019; Galligan et al., 2020; New York State Education Department, 2023).

Law and policy are not mute. UNICEF (2021) requires protecting data and privacy by design. The Committee on the Rights of the Child (2021) requires free consent, minimization, and best interests. The GDPR relativizes consent before public power and protects biometrics. COPPA and the FTC (2022) limit commercial use of school data. Miao and Holmes (2023) and UNESCO (2021) mandate privacy and a human-centred approach. Regulation (EU) 2024/1689 marks central educational AI uses as high-risk. Where sources do not measure learning or “privacy solved,” this article does not invent it. The threshold of the non-delegable is pedagogical before it is contractual: childhood capture is not delegated in exchange for an unmeasured promise.

Ingeniero Mitre / Laboratorio Editorial de NEXTECH.IA

References

  1. American Academy of Pediatrics. (2026). Digital ecosystems, children, and adolescents: Policy statement. Pediatrics, 157(2), e2025075320. https://doi.org/10.1542/peds.2025-075320
  2. CNIL. (2019, 29 octobre). Expérimentation de la reconnaissance faciale dans deux lycées: la CNIL précise sa position. https://www.cnil.fr/fr/experimentation-de-la-reconnaissance-faciale-dans-deux-lycees-la-cnil-precise-sa-position
  3. Committee on the Rights of the Child. (2021). General comment No. 25 (2021) on children’s rights in relation to the digital environment (CRC/C/GC/25). United Nations. https://www.ohchr.org/en/documents/general-comments-and-recommendations/general-comment-no-25-2021-childrens-rights-relation
  4. DiGiacomo, D. K., Greenhalgh, S., and Barriage, S. (2021). How students and principals understand ClassDojo: Emerging insights. TechTrends, 66, 172–184. https://doi.org/10.1007/s11528-021-00640-6
  5. Dutta, S., Irvin, D., and Hansen, J. H. L. (2025). Exploring discrete speech units for privacy-preserving and efficient speech recognition for school-aged and preschool children. International Journal of Human-Computer Studies. https://doi.org/10.1016/j.ijhcs.2025.103460
  6. Federal Trade Commission. (s. f.). Complying with COPPA: Frequently asked questions (sección N, COPPA and Schools; sección F, Photos, videos, and audio recordings). https://www.ftc.gov/business-guidance/resources/complying-coppa-frequently-asked-questions
  7. Federal Trade Commission. (2022). Policy statement of the Federal Trade Commission on education technology. https://www.ftc.gov/system/files/ftc_gov/pdf/Policy%20Statement%20of%20the%20Federal%20Trade%20Commission%20on%20Education%20Technology.pdf
  8. Galligan, C., Rosenfeld, H., Kleinman, M., and Parthasarathy, S. (2020). Cameras in the classroom: Facial recognition technology in schools. Science, Technology, and Public Policy Program, University of Michigan. https://doi.org/10.7302/21934
  9. ICO. (2023). The Children’s code and education technologies (edtech). Information Commissioner’s Office. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/the-children-s-code-and-education-technologies-edtech/
  10. Integritetsskyddsmyndigheten. (2019). Facial recognition used to monitor the attendance of students (decisión; sanción a Skellefteå). https://www.imy.se/globalassets/dokument/beslut/facial-recognition-used-to-monitor-the-attendance-of-students.pdf
  11. Lu, A. J., Dillahunt, T. R., Marcu, G., and Ackerman, M. S. (2021). Data work in education: Enacting and negotiating care and control in teachers’ use of data-driven classroom surveillance technology. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW2), Article 452. https://doi.org/10.1145/3479596
  12. Lu, A. J., Marcu, G., Ackerman, M. S., and Dillahunt, T. R. (2021). Coding bias in the use of behavior management technologies: Uncovering socio-technical consequences of data-driven surveillance in classrooms. En Designing Interactive Systems Conference 2021 (DIS ’21). ACM. https://doi.org/10.1145/3461778.3462084
  13. Lupton, D., and Williamson, B. (2017). The datafied child: The dataveillance of children and implications for their rights. New Media & Society, 19(5), 780–794. https://doi.org/10.1177/1461444816686328
  14. Manolev, J., Sullivan, A., and Slee, R. (2019). The datafication of discipline: ClassDojo, surveillance and a performative classroom culture. Learning, Media and Technology, 44(1), 36–51. https://doi.org/10.1080/17439884.2018.1558237
  15. Manolev, J., Sullivan, A., and Tippett, N. (2024). Reshaping school discipline with metrics: An examination of teachers’ disciplinary practices with ClassDojo. British Journal of Sociology of Education, 45(7–8), 1146–1160. https://doi.org/10.1080/01425692.2024.2407874
  16. McReynolds, E., Hubbard, S., Lau, T., Saraf, A., Cakmak, M., and Roesner, F. (2017). Toys that listen: A study of parents, children, and Internet-connected toys. En Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (pp. 5197–5207). ACM. https://doi.org/10.1145/3025453.3025735
  17. Miao, F., and Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  18. New York State Education Department. (2023, 27 September). State Education Department issues determination on biometric identifying technology in schools. https://www.nysed.gov/news/2023/state-education-department-issues-determination-biometric-identifying-technology-schools
  19. Pradhan, S., Cole, R. A., and Ward, W. H. (2024). My Science Tutor (MyST)–a large corpus of children’s conversational speech. En Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) (pp. 12040–12045). ELRA and ICCL. https://aclanthology.org/2024.lrec-main.1052/
  20. Radesky, J., Chassiakos, Y. R., Ameenuddin, N., Navsaria, D., and Council on Communication and Media. (2020). Digital advertising to children. Pediatrics, 146(1), e20201681. https://doi.org/10.1542/peds.2020-1681
  21. Regulation (EU) 2016/679 del Parlamento Europeo and del Consejo, de 27 de abril de 2016 (RGPD), arts. 8 and 9; considerandos 38, 43 and 71. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679
  22. Regulation (EU) 2024/1689 del Parlamento Europeo and del Consejo (Artificial Intelligence Act), Anexo III, punto 3 (educación and formación profesional). https://artificialintelligenceact.eu/annex/3/
  23. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  24. UNICEF. (2021). Policy guidance on AI for children 2.0. UNICEF Office of Global Insight and Policy. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children