Politics

Copy article

AI creates new dangers for children, new research finds

ended 30. March 2026

The LSE Digital Futures for Children centre published research this week making a case that deserves more attention than it will probably get. Their finding is that AI creates new dangers for children, and it makes existing dangers cheaper, faster, more personalised, and harder to detect. The difference matters, because it changes what needs fixing. There are design and accountability problems baked in before a single child ever opens the app.

The Internet Watch Foundation is already reporting rapid growth in AI-generated child sexual abuse material that evades existing detection. AI-driven tools enable the non-consensual "nudification" of images at scale. The Children's Commissioner found it is pushing girls out of online spaces entirely. Meanwhile, the UK's Online Safety Act focuses on removing illegal content after publication, not on holding platforms accountable for deploying the features that enable harm in the first place. The EU AI Act explicitly names children as a protected group. UK legislation doesn't.

The researchers also flag a less-discussed risk: over-blocking. Safety measures have been used to restrict access to LGBTQIA+ content and political information, denying at-risk young people the quality resources they specifically need, leaving them with TikTok.

Children are not edge cases in AI deployment. They are active, daily users of systems built without them in mind, and the governance frameworks are running years behind.

We'd like your views:

  • The LSE audit found AI tools already deployed in UK schools had systematic privacy violations and unsafe chatbot responses, with no evidence of child rights impact assessments by any of the companies involved. Should deploying AI in educational settings require mandatory pre-approval, as with clinical trials, or does that create barriers that entrench the advantage of well-resourced schools that can wait?
  • The Online Safety Act targets illegal content after publication. The LSE researchers argue the accountability gap is upstream: in the decision to build and deploy features that enable harm. Who should own that liability, the developer, the platform deploying it, or the institution that chose to use it?
  • AI emotional support tools are already embedded in platforms teenagers use daily. The research shows some children turn to chatbots specifically during periods of vulnerability. If the profit model rewards extended engagement, what governance mechanism actually changes that incentive, and who enforces it?

4 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
The Education Secretary called AI "the biggest boost for education in the last 500 years." The same week, LSE researchers audited generative AI tools already running in UK schools and found not one had conducted a child rights impact assessment. Sadly, that gap is the biggest factor to consider.
Copy

Editor
Dil Gujral
Chief AI Trainer at AI Now Academy
As with any technology there are benefits and downside.

The problem with AI is we are moving to a new type of technology, which has it's own challenges, problems and issues.

We only just getting our heads the dangers of social media and screen time.

No body is getting their heads around AI. Politicians in particular similar to be chasing the new shining penny of AI technology without actually understanding the dangers.

Copy

Premium
Jennie Prewett
Founder and Trustee at Incredible Kids
Teens are often ahead of their parents in their understanding of using AI and evading internet safeguards. As a result we see immense harm done to teens by unrestricted access before they are emotionally ready. As a tool, it's life-changing for young people, but AI safety and the internet in general is a big concern.
Copy

Rohit Parmar-Mistry
Founder at Pattrn Data
We keep treating children as a special interest group in tech policy. They are not. They are the default users of these systems, and AI makes the worst risks cheaper, faster, and harder to spot, long before anyone moderates a single post.

On schools: I would not copy clinical trials wholesale, but AI in education should require a mandatory pre-deployment clearance. Not a 200 page tick-box, a short child rights and safety assessment, evidence of testing, data minimisation, and a named accountable owner. Otherwise the most stretched schools become the test bed.

On liability: it has to be shared, but weighted upstream. Developers own model and safety design. Platforms own feature deployment and incentives. Schools and councils own procurement choices.

For teen wellbeing bots, the lever is the business model. If engagement drives revenue, safety will always lose. Regulators should force auditable duty-of-care metrics and real penalties, plus independent access to test systems.