AI Safety & Risk: What You Need to Know Right Now
By Saghir Hussain ·
The government's AI Safety Institute has a £100 million budget and a glossy website. What it doesn't have is a single enforcement power over the companies actually deploying dangerous systems into British homes and high streets right now.
The risks already hitting your doorstep
Forget killer robots. The real AI crisis in Britain is mundane, profitable, and already here.
Last month, a Manchester pensioner lost £12,000 to a voice-clone scam that perfectly mimicked her grandson. The fraudster needed three seconds of audio from a TikTok video. Ofcom says deepfake fraud reports have tripled since 2022. The banks? They'll tell you it's "authorised push payment fraud" and wash their hands. The telecoms providers still let spoofed numbers through. The AI companies who built the cloning tools — ElevenLabs, OpenAI, Meta — face zero liability.
Meanwhile, DWP algorithms are still sanctioning claimants based on opaque risk scores. The Information Commissioner's Office found "significant shortcomings" in how the department explains automated decisions. People in Blackpool and Birmingham are losing benefits because a model they can't see flagged them as "non-compliant." No appeal works against maths you're not allowed to inspect.
Who wrote the rules? The usual suspects
Rishi Sunak's AI Safety Summit at Bletchley Park was a masterclass in regulatory capture. The guest list read like a venture capital portfolio: Anthropic, DeepMind, OpenAI, Microsoft. Civil society groups got observer seats. Trade unions weren't in the room.
The resulting "Bletchley Declaration" is 1,300 words of voluntary commitments. No fines. No audit requirements. No right to explanation for citizens. Just "we'll try our best" from companies whose business model depends on moving fast and breaking things.
Michelle Donelan, then Science Secretary, called it a "landmark achievement." She didn't mention that her department's own white paper explicitly rules out new legislation for "the foreseeable future." The Online Safety Act — years late, watered down by lobbying — still doesn't cover AI-generated disinformation at scale.
Small businesses are the crash test dummies
A bakery in Leeds loses 40% of online orders because Google's algorithm changed overnight. A Glasgow freelancer gets deplatformed by Upwork's "quality score" with no human review. A Bristol agency watches clients vanish to £5 Midjourney subscriptions.
These aren't edge cases. They're the daily reality for the 5.5 million UK small businesses that employ 16 million people. The Competition and Markets Authority launched a foundation models review in May 2023. We're still waiting for interim findings. By the time they report, the market will be sewn up.
The government's answer? A "pro-innovation" framework that lets regulators use existing powers — powers they've had for years and haven't used. The ICO has issued exactly one enforcement notice against an AI company. One.
The bill comes due
Labour says they'll introduce binding rules for "frontier models." Keir Starmer won't define "frontier." Peter Kyle, Shadow Science Secretary, says they'll "work with industry." That's the same industry that spent £8.4 million lobbying Westminster last year.
Here's what nobody in power will say: the risks aren't theoretical. They're financial, they're discriminatory, and they're concentrated on people who can't afford lobbyists.
The morning briefing is simple. Your money, your benefits, your livelihood — they're already being decided by systems nobody accountable understands. The question isn't whether we regulate. It's whether we do it before or after the next Post Office scandal.