The UK government’s plan to ban under-16s from social media comes with a mechanism most people haven’t clocked: facial recognition and digital ID verification for every user, regardless of age. At the same time, the AI systems being trusted to enforce this kind of policy are producing hallucinated facts in government reports and consultancy papers, raising serious questions about who is really in control, and of what.
When Child Safety Becomes the Reason for Mass Surveillance
A child protection policy. That is how it has been presented. Keir Starmer’s announcement of a social media ban for under-16s, described internally as an “Australia plus” approach and intended to be law before Christmas, has been framed in the language of safeguarding. Protect the children. Shield them from harm. Who could argue with that?
The question worth sitting with is not whether children deserve protection online. Of course they do. The question is what gets built in the process of protecting them, and who it ultimately serves.
The Mechanism Behind the Measure
The ban, which the government hopes to pass through Parliament before Christmas and enforce from Spring 2027, will apply to platforms including X, Snapchat, TikTok, YouTube, gaming, and almost any website that allows for interaction. To verify user ages, the government plans to use age-recognition facial scans and digital IDs.
Read that again slowly. To access a social media platform, game or almost any website, you will submit your face.
Crucially, this process will apply to everyone, not just under-16s. The biometric verification system required to filter out minors must, by design, process every user.
This is the part that gets quietly glossed over in the coverage. The policy is not a children’s measure that adults opt out of. It is a universal biometric infrastructure built on the justification of protecting children. The stated goal and the actual architecture are not the same thing.
How Accurate Is the Technology?
Even setting aside the broader questions of consent and civil liberties, the technical picture warrants scrutiny.
Facial recognition accuracy in real-world settings, such as sporting venues, has been recorded at between 36 and 87 per cent, depending on camera placement, a significant distance from the controlled laboratory conditions under which headline accuracy figures are typically generated.
This matters. A system operating at 36 per cent accuracy in a live environment is not a security measure. It is a lottery. And yet it is being proposed as the gatekeeper to public online life.
Social media platforms will also use AI programmes to estimate user ages through facial analysis. These systems carry inbuilt biases and are prone to generating outputs that are false or fabricated yet presented with confidence, a phenomenon known as hallucination. A 2025 study found that 45 per cent of AI queries about news and current affairs produced erroneous answers, with major platforms including ChatGPT, Copilot and Gemini regularly generating factual inaccuracies.
The system being asked to verify your identity and judge your age is the same class of system that cannot reliably tell you what happened in the news last week.
When the Institutions Get It Wrong
A timely case study arrived this month from the corporate world.
KPMG’s October 2025 report on agentic AI was found, following a forensic review by research company GPTZero, to contain only five correctly cited sources out of 45. The remainder ranged from misleading and mangled to partially fabricated or too vague to verify. GPTZero estimated that roughly half of the report’s factual claims were false, unsupported, or attributed to the wrong source.
Among the specific errors: the report claimed Emirates airline had adopted a mobile chatbot named Sara that could alter flight bookings. In reality, Sara is a physical robot assistant introduced in 2023 with no ability to change bookings.
KPMG has since removed the report from some of its websites. A spokesperson confirmed the company is reviewing the circumstances of its publication and reminded staff that guidelines on responsible AI use require human oversight to validate content.
The same month, separately, it emerged that Deloitte had already refunded the Australian government after AI-generated content slipped into a taxpayer-funded report.
These are not fringe actors. These are the firms that governments and regulators turn to for expert guidance. And they are publishing AI-generated material that, on inspection, does not hold up.
Two Stories, One Picture
Place these two developments side by side and something clarifies.
A government is preparing to deploy AI-driven biometric systems across the entire population, systems with documented accuracy problems in real-world conditions, built on a technology class that leading consultancies cannot even use reliably for a PDF report, enforced through an infrastructure that will make your face a prerequisite for participation in public digital life.
The protection-of-children framing is not incidental. It is load-bearing. It is the reason the conversation moves quickly, the reason objections feel uncomfortable to raise, the reason the policy gets through before the public has time to fully understand what is being normalised.
This is how large shifts in the architecture of daily life tend to happen. Not through declaration, but through incremental steps that each seem reasonable in isolation until you look at what has been assembled.
The question is not whether this technology can protect children. The question is what else it does, who controls it, what the data is used for, and whether any of that is reversible once it is in place.
Being awake to that question is not a conspiracy. It is sovereignty.
Source: The Exposé
Join the Conversation
When governments use child safety as the foundation for mass biometric systems, how do you personally navigate the line between genuine protection and infrastructural control? And if the AI tools enforcing these systems are demonstrably unreliable, what does that tell us about the real purpose they serve? Share your experiences and insights below.
