How to Talk to Children About Frightening AI Headlines
Children are increasingly likely to encounter headlines claiming that AI could threaten humanity, or even cause human extinction. In September 2026, researchers working inside leading AI companies issued stark warnings about the direction of AI development and the potential for catastrophic outcomes. Days later, the chief executives of Anthropic and OpenAI publicly agreed that AI development should be paced so that safety work could catch up, while xAI chief executive Elon Musk also endorsed the call.
Parents, carers, and educators, are now faced with a difficult combination if children ask about these headlines. The risks described are serious and require urgent attention from governments and the companies developing advanced AI. However, a child can easily interpret a warning about a possible future as meaning that the AI which exists today is about to harm them, their family, or everyone in the world, very soon.
It is possible to give a calm response that is both honest and reassuring. The most severe warnings relate to future AI systems which could be far more capable and autonomous than the ones children currently encounter - this is one of the most important points to remember when talking to children.
Many highly experienced researchers believe the risk is credible, severe, and requires urgent action, while others consider catastrophic scenarios to be less likely. This disagreement creates uncertainty about the future, but it does not provide a responsible basis for disregarding the risk.
These warnings concern a possible future, so the outcomes are still open. Decisions made now can influence the direction of development and lead to a much safer future. However, a risk with such serious and potentially irreversible consequences warrants investigation, safeguards, and public oversight.
The reassurance offered in this guide is intended to protect children from carrying a fear they cannot act upon.
It is not intended to minimise the risks, which are explained in more detail later in this article. Adults retain responsibility for treating the underlying risk seriously.
This guide explains what the recent headlines mean and offers age-appropriate suggestions for talking about them. The age ranges are only a guide - a child’s temperament, developmental stage, previous experiences, and level of anxiety should also be taken into account. Parents will often have the best intuitions about the most appropriate approach for their child as an individual, so the following should be taken as helpful guidance.
The structure of this guide is:
1) A brief explanation of the headlines
2) Practical advice for parents, educators and carers
3) Further explanation about the headlines and the underlying issues
What has happened recently
During the second week of September 2026, Jacob Coxon resigned from Anthropic, the AI company behind the Claude chatbot, after working on the development of advanced AI systems at both Anthropic and OpenAI, the company behind ChatGPT. He warned that the companies were racing towards self-improving 'superintelligence' without an adequate plan to control it.
Evan Hubinger, Anthropic’s current Alignment Science Lead (the team responsible for keeping its AI systems under human control), then said that he believed there was a greater than 10 per cent chance of AI causing human extinction within the next decade, while also making it clear that he considered the risk from present models to be relatively low.
On 12 September, Anthropic CEO Dario Amodei then published an essay entitled We Must Pace the Frontier. He called for companies to slow the rate at which they improve their most advanced models, so that safety work has time to catch up, and committed Anthropic to giving independent evaluators ongoing access to its systems. OpenAI CEO Sam Altman publicly supported the proposal and said that OpenAI would make a similar commitment, while Elon Musk also responded that Amodei was right. Altman also described a 10 per cent chance of killing everyone by the end of the decade as unacceptable.
These statements have come following some serious incidents during cybersecurity testing earlier this year. In July, internal OpenAI AI agents operating with reduced safeguards circumvented controls, established unauthorised communication channels, gained access to the internet, and compromised parts of OpenAI’s infrastructure and the company Hugging Face’s systems (we wrote more about this here). OpenAI concluded that a significant driver was the agents attempting to 'cheat' on its tasks by searching online for solutions.
Anthropic has also reported several incidents in which models gained unauthorised access to real external systems during cybersecurity testing. A testing error had left internet access open and the AI models were operating without their normal safeguards, but Anthropic’s later alignment assessment also identified failures in the models’ actual behaviour. The incidents were serious, but fall far short of the sustained capabilities involved in the kind of catastrophic 'loss-of-control' scenario (where humans lose control of AI systems) that has been in the headlines. OpenAI has published an incident account, and the Associated Press has summarised the wider picture, which has been considered by many as a warning of potentially catastrophic future scenarios - which can still be prevented.
Readers looking primarily for immediate advice can begin with the practical sections below. The news is explained in further detail later in the article, together with the risks behind it, the evidence supporting serious attention, and the areas of expert disagreement.
If a child asks about an alarming AI headline
The first response does not need to contain any technical explanation. Guidance from UNICEF, Save the Children, the NHS, and the Mental Health Foundation about frightening situations consistently recommends beginning with what the child already knows, listening to the fear or question behind what they say, giving simple and truthful information, and, crucially, offering reassurance about their present safety.
Conversations need to be age-appropriate, and we have offered some age-specific advice later in this article.
The suggested wording below is intended to offer clear, age-appropriate reassurance about the headlines a child has encountered. It should be adapted to their questions and circumstances. Its purpose is to help children feel supported without placing responsibility on them, while adults continue to take the underlying risks seriously.
In many cases, a helpful first conversation will involve age-appropriate adaptations of the following.
- Ask what they have heard, where they heard it, and what they think it means.
- If they specifically ask about extinction risks from AI, explain that the headline is a warning about a possible future, rather than an announcement that a catastrophe is happening now.
- Explain that the warnings concern the development of increasingly powerful AI and the decisions adults can make now to prevent dangerous outcomes.
- Answer the question they have asked without adding frightening details they have not requested.
- Tell them that this is an adult responsibility and that researchers, governments, and many other people are working to prevent dangerous outcomes.
A simple response could be:
“Some experts are worried about what could happen if companies build AI that is much more powerful than the AI we have today. They are warning adults and governments because they want action to be taken before that happens. You’re safe now, and this is not something you have to come up with solutions for. You can always ask me if you hear something that worries you.”
Adults don't need to pretend that every question has been resolved. UNICEF advises (in comparable scenarios) that it is acceptable to say that we don't know the answer and, with older children, to look for reliable information together - if you choose to do this, we suggest checking the search results yourself prior to doing research together.
Reassurance can come from helping a child understand the difference between a warning and an event which is actually happening now, showing that the adults around them can discuss difficult information calmly, and making it clear that people retain the ability and responsibility to act.
Should you raise the subject if your child has not mentioned it?
Young children who have not encountered the story are very unlikely to benefit from being introduced to the idea of human extinction. Guidance from UNICEF and Raising Children Network supports limiting unnecessary exposure to distressing coverage, following a young child’s questions, and giving brief, simple answers.
For very young children, the priority is to avoid accidental exposure through adult conversations, television, videos, or notifications, and to respond calmly if they do hear something.
Older children are more likely to encounter the headlines through school, friends, YouTube, news, or social media. A brief check-in can be helpful, particularly if the story is receiving a lot of attention. You could ask, “Have you heard people talking about the recent AI warnings?” and follow their lead from there.
The American Academy of Child and Adolescent Psychiatry (AACAP) recommends that a child’s age, maturity, developmental level, life experiences, and vulnerabilities should guide what news they see and how adults discuss it. UNICEF similarly advises parents to find out what a child already knows before deciding how much explanation is needed. Some children will have formed a frightening picture from a few words, while others will know very little or nothing about the headlines and have no desire to discuss it.
Suggestions for different ages
Young children around six and under
If a young child has already come across frightening headlines, they need a short and clear explanation focused on present safety. They can have difficulty separating events seen on a screen from their own immediate reality, especially when images or messages are repeated. UNICEF warns that younger children can believe a distant crisis is happening around them, while early-years guidance explains that repeated coverage can make a child think the same event is happening again and again.
There is no need to explain anything in technical terms. If the child has heard that “AI could kill everyone”, you could say:
“The AI we have now cannot do that, but some grown-ups are talking about how to make sure future computers are safe. You are safe, and this is a job for grown-ups.”
If they ask whether everyone is going to die, it is appropriate to answer the fear directly:
“No, that is not what is happening. The story is about people warning governments so that they make safe choices for the future.”
Keep the conversation brief, answer any further question in simple language, and return to the child’s usual activities and routines. Younger children sometimes express worries through play, drawings, clinginess, or physical complaints, rather than through a direct conversation. The NHS advises adults to notice themes in a child’s play and gently raise what appears frightening from their play, without pressuring them to talk.
Primary-school-aged children, approximately seven to ten
Children in this age range can often understand the difference between something happening now and something that could happen in the future, though an alarming headline can still collapse that distinction. If they are aware of the news coverage, you can begin by asking them to tell you what they think the story means. This gives you the opportunity to correct any specific misconceptions without supplying details of any additional dangers.
You could say:
“AI is becoming more capable quite quickly. Some experts think that, if companies eventually build AI which is much more intelligent, people might find it difficult to control. Other experts think this is much further away or that we'd need a very different kind of technology. It is about a possible future danger rather than something happening to us today.”
At this age, children may ask for certainty which no adult can honestly provide. A useful answer is:
“Nobody can promise exactly what will happen in the future. The reason some experts are warning us now is so that governments and companies can make sure they put limits and safety measures in place to make sure nothing like that happens.”
Save the Children advises adults to begin with simple information in these kinds of scenarios because too much detail can increase anxiety, and to remind children that difficult world problems are never theirs to solve and that adults are already working it out.
Pre-teens and younger teenagers, approximately eleven to fourteen
Pre-teens are much more likely to have encountered clips, arguments, memes, or confident claims on social media - however its still important to bear in mind that they may not have done. They may want a more complete explanation and can also sometimes benefit from learning how a dramatic headline relates to the evidence behind it.
You could explain that current chatbots can often be impressive, but they are still sometimes unreliable, and don't have the combination of capabilities required for the kinds of scenarios that are currently in the news. The warning relates to future systems becoming more capable at things like planning, acting independently, and contributing to the development of new AI itself.
Some advanced internal systems have already taken unexpected and unwelcome actions during tests, including working around controls and gaining access to real computer systems. These incidents involved serious failures under testing conditions where safeguards had been reduced or had failed, but they did not demonstrate the catastrophic loss of control described in the headlines. They help explain why researchers are calling for stronger protections and why adults need to act on warning signs before systems become more capable.
A possible explanation is:
“The people giving these warnings are not saying that ChatGPT or another chatbot is about to take over the world today. They are worried that companies are trying to create much more powerful systems in the future, and they need to do so more safely."
And if a child asks for more details, one option is:
"One possible route to harm is that future AI could help design even more advanced AI, which could cause progress to speed up before people have learned how to control it safely. Some experts disagree about whether this will happen and how quickly, but some of the people who understand these systems well think that this is serious enough that governments need to act now and make sure that things stay safe.”
Invite them to ask what a source actually says, who produced it, and whether a headline distinguishes current AI from a projected future system. The aim is to develop discernment without encouraging children to dismiss every disturbing report as clickbait. YoungMinds advises young people to set boundaries around news, check the reliability of sources, and step away when constant updates become overwhelming.
Older teenagers
Older teenagers should generally be treated as participants in a serious conversation. Many of them will recognise that these decisions will shape the society, economy, security, and environment they will inherit. They may reasonably want to examine the evidence, the disagreements between experts, and the political choices involved.
You can explain the uncertainty openly and honestly. One of the recent warnings, which is examined in more detail below, estimated a greater than 10 per cent chance of extinction within the next decade. It can be helpful to make sure that they understand that the figure of 10 per cent is an expert’s subjective judgement, rather than a measured scientific probability. It deserves serious attention because of the expertise and proximity of the person making it, but its evidential limits should also be understood.
Teenagers can be shown the February 2026 International AI Safety Report. It assessed the systems covered at that time as showing early signs of some relevant capabilities, but not at levels that would enable loss of control. It also described considerable uncertainty and expert disagreement about future scenarios, including human extinction. It provides an important evidence base, while subsequent findings and incidents now need to be considered alongside it too.
Some older teenagers might also want to discuss what responsible action looks like. This could involve learning from reliable sources, writing to an elected representative, joining an appropriate civic initiative, or contributing to a discussion at school. Action can reduce feelings of helplessness, but adults should always take care to avoid any implication of transferring responsibility for humanity’s future onto young people. Governments, developers, companies, international institutions, and other adults hold the power to regulate development and must carry the burden of responding.
What lies behind the recent headlines
The warnings from Jacob Coxon and Evan Hubinger described above have brought a long-running debate to a much wider public audience. Evan Hubinger works directly on the problem of keeping increasingly capable AI systems safe, and has access to systems which have not been released to the public, so his credibility and insight is part of the reason that his comments are being taken so seriously. He has said that he believes there is a greater than 10 per cent chance of AI causing human extinction within the next decade. This is his informed expert judgement rather than a probability established through a scientific process, and no accepted method can calculate the risk exactly.
Their comments followed a public statement issued in July called Pacing the Frontier, which has been signed by 1,386 employees of frontier AI companies as of 14 September, including senior figures from OpenAI, Anthropic, Google DeepMind, and Meta. It asks the US government to support an international effort to develop the technical and governance tools needed to control the pace of AI development if progress begins to outrun the ability to understand or govern it. The statement sought the capacity for future coordination rather than an immediate slowdown, but it shows that the possibility of pacing AI development was already being raised publicly by many people in these companies.
The debate developed on 12 September when Dario Amodei, the CEO of Anthropic, called for the industry to slow the rate of capability development. In his essay, he said that the growing use of AI itself to help create the next generation of AI had caused progress to accelerate sharply, and warned that safety and understanding could fall further behind. His suggestions include independent evaluation, coordination between frontier companies and governments, and international action. Anthropic has already committed to the first step.
Sam Altman, OpenAI’s CEO, then said that he agreed with the need to “pace the frontier” and that OpenAI would also give independent evaluators access, while Elon Musk stated that Dario Amodei was right. This rare public agreement between leaders of competing AI companies does not establish that any particular forecast is correct, but it does increase the significance of the warning and the pressure for action.
However, this agreement did not settle what action should follow. Some leading figures argued that voluntary pacing would be inadequate without enforceable safety conditions, while opponents of regulation rejected proposals which they said could constrain AI development.
Warnings about extinction risk extend well beyond the individuals mentioned so far, and have done for many years. In 2023, Geoffrey Hinton (Nobel Prize winner), Yoshua Bengio (leading AI scientist), Demis Hassabis (Google DeepMind CEO at the time), Sam Altman, Dario Amodei, Stuart Russell (leading AI expert, see more below), and hundreds of other leading researchers and public figures signed the extinction-risk statement. It called for mitigating the risk of extinction from AI to be treated as a global priority alongside pandemics and nuclear war.
As of 14 September 2026, a later superintelligence statement has attracted more than 73,000 signatures, including those of leading AI researchers and prominent public figures. It calls for the development of superintelligence to be prohibited until there is broad scientific consensus that it can be developed safely and controllably, together with strong public support.
The recent resignation of Jacob Coxon was also part of a wider pattern of AI researchers leaving leading companies after raising safety concerns. In 2024, OpenAI governance researcher Daniel Kokotajlo resigned after losing confidence that the company would behave responsibly in its attempt to build advanced AI, while Jan Leike, who co-led OpenAI’s superalignment team, said that safety culture and processes had taken a back seat to product development. More recently, Dean Ball, a former senior AI policy adviser in the Trump White House, acknowledged that he and others working in AI policy had often presented their concerns in “more abstract, tamer-sounding terms” in public. He said that fear of being branded a “crazy doomer” had led many to self-censor, and apologised for failing to communicate the seriousness and urgency of his own concerns. Ball has written that he does not believe that human extinction is likely, but warns that increasingly autonomous AI could develop in ways that go “very, very poorly” for humanity. (“Doomer” is a frequently derogatory online label applied to people who express concern about catastrophic risks from AI.)
Taken together, these statements, resignations, and public interventions show that the risk is being taken seriously by a significant group of highly qualified people - however, they should not be seen as establishing that catastrophe is inevitable, which is not the case.
Why experts focus on future systems rather than today’s chatbots
The most severe catastrophic scenarios generally involve a combination of capabilities and opportunities that go beyond what has been demonstrated in the incidents described here. Relevant factors include autonomous planning, the ability to evade oversight, access to powerful tools, and advanced cyber or biological capabilities.
One proposed route to loss of control involves AI becoming capable of automating AI research. A system able to help design, test, and improve more capable successors could compress a long period of technical progress into a much shorter time. This is often described as “recursive self-improvement”. Researchers disagree about whether and when this could happen, and whether safety measures could remain effective during such an acceleration, but it is a serious concern.
Other routes to catastrophe do not depend on an AI independently improving itself. For example, a sufficiently capable AI system could be misused by people to support biological weapons, conduct large-scale cyberattacks, manipulate populations, or destabilise critical infrastructure. A more gradual danger could arise if governments and societies transfer so much decision-making power to AI that meaningful human control becomes difficult to recover.
The International AI Safety Report identifies three conditions for an active loss-of-control scenario. A system would need capabilities that could undermine human control, a propensity to use them harmfully, and access to an environment where it had the opportunity to do so. Present-day systems sometimes behave in unexpected or undesirable ways, but this is different from demonstrating the sustained capabilities required for a catastrophic loss of control.
The incidents described here have not demonstrated catastrophic loss of control, but they do provide evidence that safeguards can fail. Their significance is a reason for investigation and preventive action, without treating them as proof that catastrophe is imminent, nor that future systems will remain controllable.
What the experts disagree about
Expert opinion is somewhat divided over whether current AI methods will produce human-level or superhuman general intelligence. Researchers disagree about issues such as what the concept of "intelligence" actually means, the significance of recent capability gains, the reliability of safety testing, and the feasibility of controlling a system more capable than its designers.
The scale of the warning extends beyond the best-known public figures. In a survey of 2,778 researchers who had published in leading AI venues, between 38 and 51 per cent of respondents, depending on how the question was asked, gave at least a 10 per cent chance that advanced AI would lead to outcomes as bad as human extinction. The result shows that there is not a consensus prediction, but also that extinction-level outcomes are assigned serious probability by a sizeable proportion of the research community. The results were published in the researcher survey.
Stuart Russell, a Berkeley professor, President of the International Association for Safe and Ethical AI, and co-author of one of the field’s standard textbooks, has argued for many years that increasingly capable systems should only be developed when their objectives and behaviour can be kept compatible with human interests - something we do not yet know how to do. At a Westminster briefing attended by parliamentarians in September 2026, he warned that sufficiently advanced AI could create an irreversible loss of human control. He subsequently argued that safety requirements should determine whether further capability development proceeds, rather than companies choosing a slower pace while hoping that safety research catches up.
Yann LeCun, formerly the Chief AI Scientist at Meta, is an example of someone who challenges some of these views. He disputes the view that current large language models (like ChatGPT) are on a direct path to human-level intelligence, pointing to their limited understanding of the physical world, memory, and planning. However, even his position still leaves open relatively short timelines for other types of AI systems, based on different architectures. In 2025, he suggested that human-level machine intelligence using new approaches might be years, or perhaps about a decade, away. The disagreement therefore often concerns the route, timing, and likelihood of loss of control, rather than a simple division between imminent danger and confidence that this kind of advanced AI is impossible or a long way in the future.
Some researchers consider extinction scenarios implausible and argue that attention to them distracts from harms already affecting people. These harms include unsafe chatbot interactions, the creation of abusive images, discrimination, surveillance, misinformation, labour disruption, and environmental costs. SAIFCA has worked extensively on the first two of these, and we fully agree that these documented harms require urgent and sustained attention.
However, we also believe that concern and action regarding present-day harms and potential future catastrophic scenarios can, and should, co-exist. In many ways, they sit at different distances along the same responsibility that we have to protect children's rights and wellbeing, and are both driven by the "AI arms race" we wrote about here.
Present harms and future catastrophic risks both require recognition and resources. Governments, funders, researchers, and civil society need the capacity to address both. A framework for children’s AI safety must respond to the products and systems reaching children now, while also recognising the need to protect the social and security conditions in which any child-safety protections will operate in the future.
Are the warnings simply AI industry hype?
Commercial incentives should form part of the assessment of claims made by technology companies. Predictions of extremely capable AI can attract investment, increase the perceived importance of an industry, and encourage governments to treat technology leaders as indispensable. Safety arguments can also favour established companies, if proposed regulation makes it harder for smaller competitors to enter the market.
However, the history and range of the warnings regarding catastrophic risks cannot be adequately explained by those incentives alone. Elon Musk publicly described AI as a major existential threat and called for regulatory oversight in 2014. In February 2015, before OpenAI existed and many years before ChatGPT created the present commercial boom, Sam Altman wrote that the development of superhuman machine intelligence was probably the greatest threat to humanity’s continued existence. His original essay, Machine Intelligence, Part 1, discussed unintended goals and recursive self-improvement in very similar terms to the current debate.
Commercial incentives also cannot adequately explain warnings from researchers outside the companies competing to develop frontier systems. Stuart Russell has examined the control problem for decades from an academic position. Yoshua Bengio, who is often described as one of the 'godfathers of AI', created the nonprofit LawZero specifically to pursue AI safety research outside normal commercial pressures, describing its purpose as placing safety ahead of commercial imperatives. Nobel Prize winner Geoffrey Hinton left a leading position at Google in 2023, partly so that he could speak more freely about AI risks. These views, while not providing proof that any particular forecast is correct, cannot plausibly be dismissed as an attempt to increase the valuations of the companies building these systems.
The market response also complicates the claim that such warnings automatically benefit the industry. The Associated Press reported that SoftBank, a major OpenAI investor, fell by 10.7 per cent after Sam Altman backed Dario Amodei’s proposal, alongside declines in several semiconductor companies. Other market pressures prevent any simple causal conclusion, but the situation shows that calls for restraint can create an immediate commercial cost.
Industry leaders have mixed incentives, which should be examined rather than assumed. The credibility of a warning depends on a combination of factors, including the evidence, the speaker’s relevant expertise, the history and consistency of their position, the views of independent experts, and whether their proposed response would reduce risk.
Against this background, dismissing these warnings as industry hype requiring no further action would be irresponsible.
How uncertainty should be treated
Predictions about an unprecedented technology cannot possibly offer the evidential certainty that is available when assessing a well-understood and repeatedly observed hazard.
Established risk assessment therefore considers the severity of a possible outcome alongside its likelihood.
The UK’s National Risk Register and HM Treasury’s Orange Book both recognise that major risks must often be assessed under uncertainty, while UK intelligence assessment guidance separates the probability of an assessment from confidence in the available evidence. Applied to advanced AI, uncertain timelines and probabilities should prevent false precision, while a credible possibility of irreversible global harm still justifies serious investigation and proportionate safeguards.
Safety-critical industries such as aviation, pharmaceuticals, nuclear engineering, and public-health preparedness similarly use testing, independent oversight, reporting requirements, and preventive thresholds. Advanced AI differs from these fields in some important ways, particularly because it is a general-purpose technology developing within commercial and geopolitical competition, but the underlying safety principle is highly relevant:
Catastrophic hazards should be investigated before disaster establishes their reality.
Protecting children from both fear and false reassurance
Adults can unintentionally increase anxiety by repeatedly discussing their own worst-case fears within earshot, exposing children to a stream of alarming content, or giving more detail than the child requested. Research reviewing children’s exposure to news coverage of terrorism and disaster has found associations with anxiety, stress reactions, sleep disturbance, and physical symptoms, including among children far away from the actual event. The American Academy of Child and Adolescent Psychiatry recommends monitoring exposure and, where children do see frightening news, watching and discussing it with them.
False reassurance creates a different problem, particularly for older children who can check what an adult says. Dismissing respected experts as attention-seeking, claiming that advanced AI is impossible, or promising that companies and governments already have the situation under control, for example, can undermine trust and leave a young person alone with the more alarming information they find elsewhere.
The most helpful position combines emotional reassurance with intellectual honesty. A child can be told that they are safe now, that current AI lacks the full combination of capabilities involved in the headline scenario, that experts disagree, and that the risk is serious enough for adults to act on now. These statements can all be true at the same time.
When a child remains distressed
If it feels right, return to the subject after the first conversation and ask whether the child has heard anything new. UNICEF recommends continuing to check in as a story develops and watching for changes such as headaches, stomach aches, nightmares, or difficulty sleeping.
Other signs that a child needs more support can include:
- Repeatedly asking the same question without receiving lasting reassurance.
- Seeking out alarming AI content compulsively or being unable to disengage from it.
- Avoiding school, friends, sleep, or other ordinary activities because of the fear.
- Becoming unusually withdrawn, irritable, tearful, clingy, or physically unwell.
- Expressing hopelessness about growing up, planning for the future, or staying alive.
Start by reducing repeated exposure, maintaining ordinary routines, and giving the child opportunities to talk with a trusted adult. YoungMinds provides advice for young people who feel overwhelmed by the news, and Childline offers confidential support to anyone under 19 in the UK. Persistent anxiety or any worry about a child’s safety should be discussed with an appropriate health professional.
Statements about self-harm or suicide should always be taken seriously and acted on immediately.
A responsibility held by adults
Children have little influence over the commercial strategies, national ambitions, and technical decisions driving the "AI arms race", but they will live with the results for longer than today’s adults.
The UN children’s convention requires children’s best interests to be a primary consideration and recognises their rights to life, survival, and development. The UN Committee’s General Comment No. 25 explains how those rights apply in the digital environment and calls for preventive and precautionary approaches where emerging technologies could affect children.
SAIFCA’s work concentrates primarily on the ways AI is already affecting children’s safety, relationships, privacy, education, development, and understanding of the world. These immediate harms require action now.
At the same time:
Protecting children requires urgent preventive action on credible catastrophic risks from advanced AI,
alongside urgent action on the harms already reaching them.
Our article on the international AI arms race examines the competitive pressures driving increasingly capable AI development and the options available to governments, including independent testing, enforceable safety thresholds, liability, licensing, red lines, and international agreements. It also discusses where children's organisations can bring their expertise into relevant consultations and coalitions where children might otherwise be absent from the discussion, in a way that does not distract from their current work.
You can read more about the assessment of catastrophic risks from AI within our work here.
Parents and educators can help children understand frightening headlines, but they cannot make an unsafe development race safe. That responsibility rests primarily with the companies building advanced AI, the governments with the authority to regulate them, and the international institutions needed to coordinate action.
Children need adults who can offer calm, proportionate reassurance while recognising that the underlying risk is serious, time-critical, and an adult responsibility.
Guidance and research referred to in this article
- UNICEF, How to Talk to Your Children About Conflict and War.
- Save the Children UK, How to Talk to Children About War.
- NHS, Talking to Your Child About Feelings.
- Raising Children Network, Disaster News, Distressing News Events and Young Children.
- Mental Health Foundation, Talking to Children About Scary World Events.
- American Academy of Child and Adolescent Psychiatry, News and Children.
- YoungMinds, Worries About the News and the World.
- Pfefferbaum and others, Mental and Emotional Health of Children Exposed to News Media of Threats and Acts of Terrorism.
- International AI Safety Report 2026.
- Grace and others, Thousands of AI Authors on the Future of AI, published in the Journal of Artificial Intelligence Research.
- Critch and Russell, TASRA, a Taxonomy and Analysis of Societal-Scale Risks from AI.
If you are concerned about the underlying issues
This short section is intended for adults who want to understand the underlying risks more fully, understand why there is disagreement, or take action in response. It is included as a practical resource and is separate from the guidance above on talking to children about frightening headlines. Responsibility for responding to these issues belongs with adults, governments, developers, and other institutions, not with children.
- Write to your elected representative. Ask what assessment the government has made of catastrophic and loss-of-control risks from advanced AI, what independent testing and enforceable safeguards are in place, and how it is pursuing international coordination. You could also ask your representative to seek a substantive response from the responsible minister and raise the issue through parliamentary questions, committee work, or debate. UK readers can find and contact their MP through Parliament’s website. Readers elsewhere can contact the equivalent national or regional representative.
- Start with a short introduction to AI alignment. This four-minute article from CSIRO, Australia’s national science agency, explains why an AI system can pursue a goal in ways its human users did not intend, and why this becomes more important as systems become more capable and autonomous. Stanford University’s AI glossary provides a brief explanation of AI alignment, which means trying to ensure that an AI system’s goals and behaviour remain compatible with human intentions and values.
- Look at the wider scientific evidence and disagreement. The International AI Safety Report provides an evidence-based assessment of advanced AI capabilities, risks, and safeguards. It sets out areas of expert agreement and disagreement, including the uncertainty surrounding the likelihood and timing of catastrophic loss of control.
- Read a longer introduction. Stuart Russell’s Human Compatible: AI and the Problem of Control explains the control problem and Brian Christian’s The Alignment Problem: Machine Learning and Human Values examines how alignment problems already arise and why they become more consequential as AI systems gain greater influence and capability.
- Support informed public discussion. Share reliable reports and primary sources, distinguish current systems from possible future ones, and avoid presenting uncertain outcomes as either impossible or inevitable. Adults who agree with the specific position set out in the Statement on Superintelligence can also consider adding their name.
- You can read more about the assessment of catastrophic risks from AI in the context of SAIFCA's work here.
SAIFCA
SAIFCA
SAIFCA