Britain’s escalating social‑media clampdown is being sold as child protection. Look closer at the technology being deployed, and it looks much more like a project to hard‑code political speech controls into the digital infrastructure that now mediates almost every conversation we have.
As was argued previously, free speech in the UK is no longer primarily threatened by explicit bans. It is being eroded by a stack of AI moderation systems, pre‑emptive scanning mandates and open‑ended powers for ministers and regulators to decide, in practice, what we are allowed to say online. Westminster and Holyrood are both pulling on that lever – and the tech sector is where those decisions become real.
From “safety duties” to real‑time AI scanning
The Online Safety Act is the backbone of this new regime. On paper, it imposes duties on platforms to tackle illegal content and protect children, including content generated or amplified by AI systems. In practice, it explicitly anticipates that providers will rely on automated systems to decide what stays up and what is suppressed.
Amendments taking effect this year push that logic to its limit. On 8 January 2026, the UK expanded the Act to require “real‑time, AI‑driven content scanning across messaging apps, forums, and search engines” for certain categories of material, including cyber‑flashing and encouragement of self‑harm. Platforms must “take proactive steps to prevent this vile content before users see it,” shifting the model from post‑hoc moderation to pre‑emptive interception. Non‑compliance can mean fines of up to 10% of global turnover or £18m, and even service blocking.
Technically, that means every message, image or clip on covered services passes through machine filters before it reaches its recipient. Architecturally, it means the UK has normalised a system where private communications are always subject to algorithmic scrutiny, and where the criteria for what must be stopped can be widened later by secondary legislation. Once that infrastructure is in place, almost anything governments choose to label “harmful” can be slotted into the same pipeline.
This is exactly the kind of mechanism an over‑zealous state would choose if it wanted to dial lawful but disfavoured speech up or down without ever announcing a formal ban.
Scotland: law as a censorship force multiplier
Nowhere is the political appetite to use that machinery more obvious than in Scotland. The Hate Crime and Public Order (Scotland) Act’s “stirring up hatred” provisions are so vague that almost any forthright opinion on race, religion, gender identity or national identity can be portrayed as suspect. The result, as we’ve already seen, is thousands of complaints pouring in within days of the law coming into force, including against the then First Minister’s own speech.
On their own, those complaints would be a policing headache. Combined with AI moderation and UK‑wide duties to suppress “potentially illegal” or “harmful” content, they become something else: a stream of enforcement signals that can be fed into platforms’ risk engines and trust‑and‑safety dashboards. The more expansive Holyrood is in treating contested opinions as hate, the more justification regulators and platforms have to down‑rank or remove that speech pre‑emptively “just in case”.
In other words, Scottish law is not just narrowing what can be said on a stage or in a newspaper column. It is supplying a constant flow of labelled “bad” content that will inevitably train the next generation of moderation models.
AI chatbots dragged under the same umbrella
The government is now extending this framework to generative AI itself. After Ofcom admitted it lacked tools to address Elon Musk’s Grok chatbot being used to generate non‑consensual sexualised images, ministers pledged to “move fast to shut a legal loophole and force all AI chatbot providers to abide by illegal content duties in the Online Safety Act.” Keir Starmer has since announced that chatbots will be brought under the same illegal‑content rules as social‑media platforms, via amendments to wider crime and policing legislation.
Again, the headline aim – tackling AI‑generated abuse – is hard to oppose. But the mechanism is familiar: pull a whole class of technologies (chatbots, LLM‑powered tools) into a regime where senior ministers will have “power to amend any provision of the Online Safety Act in relation to illegal AI‑generated content.” In code, that translates to one thing: yet more centralised control over what models are allowed to generate or even discuss, enforced by threat of ruinous fines.
Technically, we are building a world in which both human‑to‑human and human‑to‑AI conversations are mediated by systems whose primary design constraint is regulatory fear.
Platforms in the dock: real harms, bigger leverage
The political window for all this has been opened by genuine, court‑recognised harms. In Los Angeles, a jury recently found Meta and Google negligent over the design of Instagram and YouTube, ruling that the platforms were “intentionally crafted to be addictive” and awarding a young woman US$6m in damages. Jurors heard that internal communications compared their apps to drugs and gambling, and that “viewer addiction” was an explicit product goal.
That verdict is likely to unleash a wave of litigation and force design changes worldwide. It also hands governments a powerful talking point: if juries are concluding that social platforms are effectively defective products, then aggressive regulation of feeds, features and recommendation algorithms can be portrayed not just as permissible, but morally urgent.
The question is what sort of regulation follows.
Tech architecture, not just “culture war”
It is tempting to view all this as just another left–right culture war. But for a tech‑savvy audience, the more important story is architectural. We are deciding, right now, whether the default design of communications systems in the UK will be:
- end‑to‑end private by default, with targeted, warrant‑based interception in exceptional cases,
or
- always‑on, AI‑mediated scanning with ever‑expanding lists of “harmful” categories feeding the filters.
Scottish legislation, with its low bar for criminalising “stirring up hatred”, and Westminster’s broad “illegal and harmful” duties under the Online Safety Act, are not separate issues. They are two layers of the same stack: norms and definitions at the top, automated enforcement at the bottom.
For developers, founders and engineers, this should be a first‑order concern. Pre‑emptive scanning mandates and open‑ended illegal‑content duties have direct implications for:
- the cost and complexity of building compliant products,
- the viability of end‑to‑end encryption,
- the feasibility of hosting controversial but lawful speech, and
- the ability of UK‑based companies to compete with services in jurisdictions that have chosen a lighter‑touch model.
The choice facing the tech community
None of this is an argument for ignoring real harms, or for pretending the California verdict is a fluke. Platforms that knowingly design “digital casinos” for children deserve the legal trouble now coming their way. But if we care about both safety and a free internet, we cannot leave the design brief entirely in the hands of politicians who see speech primarily as something to be managed.
The UK – and Scotland in its own right – is on the cusp of embedding a model in which speech rights are contingent on what AI filters, regulators and ministers deem acceptable at any given moment. Tech companies, engineers and digital‑rights advocates have a narrow window to insist on a different path: one that targets specific, evidenced harms in narrowly coded ways, rather than handing the state a general‑purpose dial on the underlying machinery of expression.
We need to look deeper than the darkness of the planned social‑media clampdown. Its about more than who gets banned this week, who gets arrested for a social media post, it’s about what kind of communications infrastructure are we building for the next decade and where that leaves our ever diminishing freedom of speech.
