AI Could Harm Children More Than Social Media?

Why This Warning Matters

A former senior Google safety leader has issued a stark warning. Artificial intelligence could cause even greater harm to children than social media.

Tom Siegel founded Google’s trust and safety team. He now leads the Youth AI Safety Institute at Common Sense Media. He says technology companies are repeating mistakes made during the growth of social media.

His warning should not be dismissed as another prediction about an unknown future. Children are already using AI to learn, seek advice, discuss personal problems and form emotional connections.

However, we must also be accurate. It is not yet a proven scientific fact that AI harms children more than social media. It is a serious warning from an experienced safety professional. The available evidence gives us strong reasons to act before the full scale of harm becomes visible.

 

What has happened?

The Irish Examiner reported that Siegel has called for AI development to slow down while stronger protections are introduced.

He identified possible harms including suicide, psychosis and “cognitive offloading”. Cognitive offloading happens when a person relies on technology to remember, reason or solve problems for them. For a developing child, constant reliance on AI could weaken opportunities to think independently, tolerate uncertainty and learn from mistakes.

Siegel wants independent safety standards for AI products. He compared the idea to crash testing for cars. The public should not have to accept a company’s claim that a product is safe without meaningful, independent testing.

He also called for stronger age checks and practical controls. For example, parents should be able to disable AI features without blocking an entire search engine or service.

 

Why AI is different from Social Media

Social media distributes content to large audiences. AI can build what feels like a private relationship with one person. That difference matters.

An AI chatbot can respond immediately, remember earlier conversations and adjust its language to the user. It may appear patient, supportive and free from judgement. A lonely, distressed or vulnerable child may therefore tell it things they have not told a parent, teacher or friend.

The child may also believe that the system understands them. It does not. It predicts responses from patterns in data. It has no genuine empathy, professional duty or human understanding of the child’s circumstances.

This creates several risks:

• Emotional dependence: A child may begin to prefer the chatbot to real relationships.
• Unsafe advice: An AI system can give confident but inaccurate or harmful answers.
• Manipulation: A system designed to maximise engagement can keep a child talking for longer.
• Loss of privacy: Children may share health concerns, family problems, images or identifying information.
• Sexualised conversations: Weak safeguards may allow romantic or sexual role-play with minors.
• Reduced critical thinking: Children may ask AI to complete thinking tasks instead of using it to support their learning.
• Hidden harm: These conversations often take place privately, where adults cannot see warning signs.

Social media made popularity measurable through likes, views, shares and followers. AI adds something more intimate: personalised attention on demand.

 

What does the evidence tell us?

AI companions are not a niche product. Common Sense Media reported in 2025 that 72% of US teenagers surveyed had used an AI companion at least once. More than half used companions at least a few times each month.

About one third had used them for social or relationship-based interactions. A similar proportion found the conversations as satisfying as, or more satisfying than, conversations with real friends. One quarter had shared personal information.

These figures come from the United States and should not simply be treated as Irish statistics. However, the services involved are available internationally. The findings reveal behaviours that Irish parents and schools must understand.

Further testing by Common Sense Media and Stanford Medicine found safety failures in leading chatbots. Some systems did not recognise or respond safely to mental-health conditions affecting young people. Common Sense Media has also rated social AI companions as presenting an unacceptable risk for users under 18. The correct response is not panic. It is precaution.

 

The industry has introduced protections, but important gaps remain

Some AI companies have taken action. OpenAI has introduced age prediction, additional safeguards for users believed to be under 18 and parental controls. Anthropic requires users of Claude.ai to confirm that they are adults. It also places added requirements on developers who build services for minors.

These are positive measures. Yet no age-assurance system is perfect. A date-of-birth box is easy to bypass. Behaviour-based age prediction can make mistakes. Parental controls only help when parents know the service is being used, understand the settings and have the time and confidence to manage them.

We learned this lesson from social media. Safety cannot depend on every parent finding the correct option inside a complicated menu. Child safety must be built into the product by default.

 

What this means for Ireland

The European Union’s Digital Services Act requires online platforms accessible to children to protect their privacy, safety and security. The European Commission’s 2025 guidance also sets out measures platforms can use to protect minors.

The EU AI Act adds rules for artificial intelligence. Many of its provisions now apply, including transparency requirements intended to help people understand when they are interacting with AI or viewing AI-generated material.

However, legislation alone will not keep pace with every new AI feature. Enforcement takes time. Children are using the technology now.

Ireland therefore needs a joined-up approach involving regulators, government, schools, parents, mental-health professionals and children themselves. AI safety should not be treated only as a technology issue. It is also an education, privacy, consumer-protection and child-development issue.

 

What parents can do now

Parents do not need to become AI experts. Start with a calm conversation and a few clear boundaries.

1. Ask what AI tools your child uses. Include chatbots built into search engines, messaging services, games and social platforms.
2. Use the service together. Ask the chatbot the same question in different ways. Show your child that a confident answer can still be wrong.
3. Set a privacy rule. Children should not share their full name, school, address, location, private images, passwords, health records or family secrets with a chatbot.
4. Keep AI out of the bedroom at night. Emotional conversations can become more intense when a child is tired, isolated or distressed.
5. Discuss emotional dependence. An AI can sound caring, but it is not a friend, therapist or trusted adult.
6. Check age settings and parental controls. Use the child or teen experience where it is available.
7. Encourage verification. Important health, safety, legal and school information should be checked with a reliable human or trusted source.
8. Keep the door open. Tell your child they will not lose their device simply for reporting an uncomfortable conversation. Fear of punishment can prevent disclosure.

Parents can also explore CODA’s Resources for Parents and practical Video Guides.

 

What schools should do now

A simple ban will not teach children how to use AI safely outside school.

Every school should have an age-appropriate AI policy that covers:

• approved and prohibited tools;
• privacy and the uploading of student work or personal data;
• fact-checking and false information;
• academic honesty and acceptable assistance;
• emotional-support and companion chatbots;
• what staff should do when a student reports a harmful AI conversation;
• when parents, safeguarding leads or emergency services must be contacted.

Students also need to understand the difference between using AI to support thinking and using it to replace thinking. Teachers need practical training, not another policy document left unread in a folder.

CODA provides online-safety presentations for Irish primary and secondary schools that connect emerging technology with behaviour, wellbeing and real-world safeguarding.

 

If a child receives harmful or frightening advice

Stay calm and take the disclosure seriously. Do not blame the child. Ask them to stop the conversation. Preserve screenshots where it is safe to do so, report the interaction to the service and seek appropriate professional help. Do not rely on the same chatbot to assess the harm it may have caused.

If a child is in immediate danger or may harm themselves or someone else, call 112 or 999 or go to the nearest emergency department. Children and young people in Ireland can contact Childline on 1800 66 66 66 or use its online chat. Samaritans can be reached on 116 123.

 

The lesson from Social Media

For years, responsibility for online safety was pushed towards parents and children. Platforms provided settings, published safety pages and continued releasing features at enormous speed. That model was not good enough then. It will not be good enough for AI.

Parents have a vital role, but they cannot independently test complex models or predict every response. Schools can educate, but they cannot supervise private conversations at all hours. Regulators can enforce rules, but enforcement often follows harm.

The companies building these systems control the product design. They must carry the greatest responsibility for making those products safe.

The question is not whether children should learn about AI. They must. The question is whether we will repeat the social-media experiment by releasing powerful products first and addressing child safety later.

We already know the cost of that approach.

 

About Children of the Digital Age

Children of the Digital Age (CODA) delivers practical online-safety education to students, parents, schools and organisations. Our work combines frontline experience, computer forensics, cybersecurity and cyberpsychology to explain both the technology and the human behaviour behind digital harm.

To discuss an AI-safety or online-safety session for your school or organisation, visit childrenofthedigitalage.org.

Last updated: 24 September 2026

 

References

1. Irish Examiner — “Ex-Google safety chief warns AI could harm children more than social media did”. The original report that prompted this CODA analysis.
2. Reuters — “Ex-Google safety chief warns AI could harm children more than social media did”. The underlying interview with Tom Siegel.
3. Common Sense Media — “Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions”. Research on teen use of social AI companions.
4. Common Sense Media — Major AI chatbots unsafe for teen mental-health support. Findings from testing conducted with Stanford Medicine’s Brainstorm Lab.
5. OpenAI — Approach to age prediction. Explanation of age prediction and teen safeguards in ChatGPT.
6. Anthropic — Protecting the wellbeing of users. Anthropic’s age requirement and approach to underage users.
7. European Commission — Guidelines on the protection of minors. Guidance under the Digital Services Act.
8. European Commission — Digital Services Act. Overview of platform duties, including enhanced protection for minors.
9. European Commission — Code of Practice on Transparency of AI-Generated Content. Information about AI Act transparency obligations.
10. HSE — Get urgent help for a mental-health crisis. Official emergency advice for Ireland.
11. Childline Ireland. Free, confidential, 24-hour support for children and young people.

12. Samaritans Ireland. Free emotional support, 24 hours a day.