There’s a very British pattern to how we absorb new technology. First, suspicion: remember the general air of doubt around contactless payments, which lasted approximately one year before we were all tapping our cards for a bus fare. Then wholesale adoption, without anyone quite pausing to ask what we’d agreed to. Bank apps, smart meters, loyalty schemes that know our weekly shop better than our spouses do: each adoption felt individually sensible, and the cumulative picture, a detailed portrait of British households held by commercial interests, drew remarkably little commentary.
The latest chapter is the biggest yet. Within roughly two years, AI chatbots went from curiosity to furniture: millions of us now run questions, drafts, plans and problems through an assistant as casually as we once checked Ceefax. Schoolchildren use them for homework, pensioners use them to decode utility bills, small businesses use them to write the flyers. The adoption question has been answered emphatically. The questioning part, what we’ve actually signed over, is where we’re behind schedule.
What we’re actually telling them
It’s worth pausing on what flows through the average household’s chat window. Salary negotiations rehearsed the night before. Symptoms typed in before booking the GP appointment. The disciplinary letter drafted at 11 p.m. Homework help that reveals exactly which child struggles with which subject. Small business plans, cash-flow worries, tax questions. A bereavement, in the middle of the night, asking how to write to the solicitor.
Typed out like that, it reads like the contents of a counselling session. And on mainstream platforms, that material is retained on company servers, under terms of service that have permitted its use for model training. The Electronic Frontier Foundation’s Surveillance Self-Defense guide documents both the retention practices and the incidents where conversations leaked or appeared to other users. British households have, in effect, been keeping an extremely frank diary in someone else’s filing cabinet.
For UK readers there’s a regulatory backdrop worth knowing. The Information Commissioner’s Office, the body that polices UK data protection law, has been actively examining exactly this territory: how personal data flows into AI systems, what constitutes lawful processing, and what questions consumers should be asking. Its guidance on AI and data protection is not bedtime reading, but its existence signals something important: the regulators already consider this a real issue. Ordinary users are just the last constituency to get briefed.
The specific British risks
Scam season, upgraded. The British consumer faces one of the more industrialised scam economies anywhere: the fake HMRC calls, the bogus courier texts, the TV Licensing emails. The National Cyber Security Centre has documented how these operations are increasingly data-driven: they work because the fraudster already knows enough about you to sound plausible. An assistant that harvests its users’ questions is, functionally, an enrichment programme for that ecosystem.
The family IT department problem. In most households, one person is the informal tech support for parents, in-laws and neighbours. The questions those relatives type into an assistant, often health, benefits, pensions, loneliness at 2 a.m., are precisely the material that should have the strongest custody. Helping older family members onto a privacy-first assistant is the modern equivalent of setting up their caller display to screen the scam calls.
Where the data lives. UK GDPR gives you rights over your data, but those rights are easier to exercise when the company holding it operates under a comparable legal regime. Where your conversations physically reside, and whose courts can demand them, matters when the questions involve your mortgage, your health or your job.
The alternative that speaks plain English
The constructive part of the catch-up: choosing a better-constructed assistant is now easy, and the strongest privacy-first options are European, which ought to please British sensibilities about both quality and proximity. A privacy-first AI assistant answers the questions above directly: no server-side logs of your conversations, so there is no master archive of your questions on the provider’s infrastructure. Encryption on anything you save, meaning chats can be decrypted only on your own devices, not by the provider. Conversations never used to train models, and never used for targeted advertising, because there’s no advertising in the business model at all; it’s funded by subscriptions. Fully open-source code, so the whole arrangement can be inspected rather than taken on trust. And a base under strict European data protection law. A genuinely usable free plan, with paid tiers for heavier use.
Crucially, it’s also simply a good assistant: drafting, explaining, planning, translating, generating images, searching the live web, built on the same class of language models as the mainstream tools. The old trade-off, capability for privacy, has quietly expired.
The British way through
There’s no need for either technophobia or techno-worship here; both are forms of not thinking. The sensible position, the one we eventually reached about chip-and-pin, about online banking, about every technology absorbed in the usual order of suspicion, adoption and belated scrutiny, is to use the tools well while asking what they’re doing with the material. Ask it of your assistant. And if the answer comes back as vaguely as a terms-of-service update, there are now assistants that answer with architecture instead: no logs, no training, encryption that holds even against their own maker.
Millions of Britons already tell an AI assistant their problems. The only real question left is whether anyone else gets to keep the transcript.
