A teenager pastes a private conversation into a chatbot, uploads a photograph to make an avatar, uses an invented answer in an assignment, or follows advice that causes harm. Then the problem becomes visible and the teenager stays silent. Silence does not always mean failure to understand the risk: they may fear losing their phone, being punished at school, being mocked at home, or being blamed by friends. Safety begins before any technical adjustment. It begins when a young person can say, «This happened», and meet someone able to listen, reduce harm, and establish responsibility without turning one bad decision into an identity.
Adolescence brings together curiosity, a need to belong, learning through trial, and particular sensitivity to social judgement. Artificial intelligence tools intensify that combination: they answer immediately, appear confidential, and allow users to test identities, relationships, and solutions without the discomfort of talking to another person. That availability can be useful, but it also makes mistakes likely. A policy that demands perfect use from the first day merely ensures that incidents will be concealed.
Trust does not remove boundaries: it makes them work#
A trusting environment does not accept every action or promise that consequences will never follow. It means that the first response separates protection, understanding, and decisions about responsibility. If an intimate image is circulating, a threat has been made, a scam may be under way, or health is at risk, the priority is to stop dissemination, preserve only essential evidence, and activate support. Education and proportionate consequences follow once the urgent danger is contained.
A trusting environment does not accept every action or promise that consequences will never follow.
Intent, impact, and context should also be separated. Deliberately sharing material to humiliate someone is not the same as entering personal data into an app through lack of knowledge. The harm may be serious in both situations, but the intervention differs. Asking «What did you expect to happen?», «What actually happened?», and «Who may be affected?» produces better information than beginning with «What were you thinking?». Precision does not excuse conduct; it supports better repair.
Trust is built before an incident. Families and schools can agree on a simple rule: asking for help early reduces harm and will be taken into account. They must honour it. If every disclosure produces public exposure, automatic punishment, or a lecture, the lesson is to conceal the next incident. An adult can maintain clear boundaries while acknowledging the decision to speak: «This matters and we will address it. First, I need to understand it with you».
Why reporting an AI failure can feel difficult#
Many interfaces create a false sense of intimacy. A system writes in the first person, remembers details, and responds without surprise. A teenager may attribute understanding, loyalty, or secrecy to it, although providers, logs, moderation systems, and terms of service sit behind the exchange. Admitting that sensitive information was disclosed can feel like acknowledging a betrayal of oneself. Shame increases when the app was prohibited or the conversation concerned sexuality, mental health, identity, family conflict, or violence.
Ambiguity is another barrier. A false answer may sound reasonable; an image filter may alter a body gradually; a recommendation may reinforce isolation without giving an obviously dangerous command. The young person may feel distress without being able to prove a «failure». Adults should accept incomplete reports. They do not need to decide immediately whether the cause was a hallucination, persuasive design, misuse, or coincidence. Recognising the effect and examining events is enough to begin.
There may also be fear of losing access. For someone whose friendships, schoolwork, or creativity depend on a platform, reporting a problem can feel like exclusion from the group. A proportionate response therefore looks for alternatives: pausing one feature, changing privacy settings or credentials, or moving to an approved tool may protect better than confiscating every device. When a full restriction is required by immediate danger, its reason, duration, and review criteria should be explained.
Three cases for learning to respond without blame#
Case 1: an emotional conversation escalates#
Lucia, aged fifteen, talks to an assistant late at night about anxiety. The system answers confidently, suggests withdrawing from people who «do not understand her», and minimises a sentence about self-harm. Lucia deletes part of the history because she fears that her parents will ban internet access. A friend alerts the school counsellor. The right response does not start by deciding whether Lucia broke a rule. It calmly and directly checks for immediate danger, does not leave her alone with an urgent risk, and contacts emergency services or health professionals when appropriate.
An adult then retains only necessary information, reviews with Lucia which messages influenced her, and explains that a chatbot is not clinical care. They agree on human support, safer times, and a simple way to ask for company. If there is risk of self-harm or suicide, no one promises absolute secrecy and immediate local help is activated. The technology is examined without assigning it a diagnosis or human intention; safety and continuity of care come first.
Case 2: an image made as a joke#
Pol, sixteen, generates a fake sexualised image of a classmate and shares it with a small group. Someone forwards it. Pol wants to disclose what he did but fears exclusion and expulsion. Asking for help does not erase the harm or make his fear the centre of the case. The adult must first protect the depicted person: prevent new copies, avoid asking her to confront the creator, preserve minimal evidence through safe channels, and activate the school safeguarding procedure. The image is not forwarded «for verification» or stored on personal devices.
The intervention with Pol requires specific accountability: acknowledge the impact, stop dissemination, cooperate with legitimate measures, and participate in repair that respects the affected person’s wishes. A direct apology is not always safe or wanted. Competent professionals should assess possible legal and disciplinary duties, particularly where minors and sexual content are involved. Education must address consent, dignity, and the ease with which a synthetic image escapes its original setting.
Case 3: an assignment containing fabricated evidence#
Amir uses a generator to prepare a science report. The tool invents two studies and he includes them without checking. When the teacher requests the sources, Amir considers maintaining the lie. A safe reporting route lets him tell the truth before the problem grows. The teacher can withdraw the grade provisionally, request a reconstruction of the process, and offer a new submission under clear criteria. The consequence relates to learning that was not demonstrated; it does not label the student a cheat forever.
Amir compares every claim with real sources, marks the passages produced with the tool, and explains how he will verify material next time. If the school policy was vague, the incident also exposes institutional responsibility: staff must clarify when AI may be used, how its use is disclosed, and which evidence of process is required. Early honesty should affect the response, even though the work still needs to be completed properly.
A short protocol for the first twenty-four hours#
The first conversation can follow five movements. First, regulate the adult response: use a calm voice, short questions, and no impulsive posting. Second, check for danger involving physical safety, extortion, sexual contact, exposed location, fraud, discrimination, or medical risk. Third, stop reversible processes: close a session, change a password, block a contact, or request content removal without destroying evidence that may be needed. Fourth, identify anyone else requiring protection. Fifth, agree on the next action and a time for another conversation.
Documentation does not mean collecting sensitive material. Record the date, platform, account, nature of the output, recipients, and actions taken. Screenshots should be limited to what is essential and held through authorised channels. Intimate images of minors should not be downloaded or redistributed; authorities or specialist services should provide guidance. Nor should a teenager be questioned repeatedly. Recounting traumatic details to several people can compound harm.
Important decisions should not depend on one individual. In a school, the safeguarding lead, counsellor, leadership team, and data protection officer may hold distinct responsibilities. At home, health services, police, social care, or legal support may be needed according to the risk. If the school supplied the system, it must record the incident, provide required notifications, and review whether its procurement, settings, or oversight contributed to the outcome.
Conversations that open a door and phrases that close it#
Helpful phrases include «Thank you for telling me», «You do not have to solve this alone», «I will ask only what I need to protect you», and «I will explain before sharing this, unless danger requires action». They promise neither impunity nor impossible confidentiality. They create predictability. Small choices also help: speaking now or in ten minutes, writing first, choosing which trusted adult is present, or deciding whether a short pause is needed.
Phrases such as «I warned you», «That is what happens when you use it», «Show me everything right now», or «If you cared, you would have told me earlier» close the door. Public humiliation, jokes, and disproportionate threats shift attention from the harm to self-protection. Even when a young person has harmed somebody else, the goal is to obtain reliable facts, protect the victim, and promote responsibility. Shame may produce short-term compliance, but it rarely develops sound judgement.
A particularly valuable question is «What do you need me not to let happen while I help?». The answer may reveal fears about retaliation, involuntary disclosure of identity, violence at home, or loss of friendships. The adult cannot always guarantee the requested outcome, but can explain what is controllable and make a safety plan. Where there are signs of abuse, exploitation, or coercion, applicable safeguarding duties should be followed and the suspected person should not be confronted informally.
Shared responsibility: the burden is not all on teenagers#
Platforms should design for real users, including foreseeable use by minors, with high privacy by default, understandable notices, boundaries against harmful content, and accessible reporting. Schools should assess tools before requiring them, minimise data, provide a human alternative, and train staff. Families need rules they can explain and revisit. Blaming only the young user hides the commercial and institutional decisions that shape what is easy, attractive, or difficult to leave.
UNICEF’s policy guidance on AI and children places wellbeing, inclusion, privacy, safety, and accountability within the design and use of systems. UNESCO’s guidance for generative AI in education and research calls for a human-centred approach, data protection, and age-appropriate use. These sources support demands for responsible conditions; they are not merely lessons teaching children to adapt to whatever technology provides.
In Spain, the Spanish Data Protection Agency’s official guidance collection covers minors, education, privacy, and risk management. The European Union Artificial Intelligence Act on EUR-Lex establishes a risk-based framework and duties for specified systems. Precise application depends on the use, provider, date, and jurisdiction; serious harm requires advice from competent professionals and authorities.
Turning an incident into learning and repair#
Closing a case requires more than disabling an account. It means checking whether harm continues, who needs information, which data should be deleted or corrected, what emotional support remains available, and which change can prevent recurrence. A brief meeting several days later can reveal further sharing, retaliation, anxiety, or new questions. Follow-up shows that requesting help was not a formality and that a young person is not abandoned after disclosure.
Repair should consider everyone affected without forcing reconciliation. It may involve removing content, correcting a claim, redoing an assignment, restoring access that was unfairly blocked, acknowledging an institutional error, or changing a configuration. A victim should not be required to educate the person who caused harm. Nor should the responsible teenager permanently become «the problem»: accepting consequences and developing new capabilities are part of prevention.
Finally, a community can learn without identifying or exposing anyone. A school can say that it has detected a category of risk, remind people of the support route, and update practice while omitting details that identify those involved. Trust grows when rules are clear, adults honour what they promised, and mistakes lead to visible improvement. A generation that talks with machines needs to know something profoundly human: reporting a mistake early does not reduce a person to that mistake; it opens the possibility of protection, accountability, and repair.
