AI Safety & Risk: What You Need to Know Right Now
By Saghir Hussain ·
AI safety isn't a theoretical debate for boffins at Bletchley. It's a trading standards issue hitting UK high streets and hospital wards today.
Last month, a Manchester pensioner lost £12,000 to a voice-clone scam replicating her grandson's WhatsApp plea. The fraudster used eleven seconds of a TikTok video. Eleven seconds. Ofcom says AI-enabled fraud cost Britons £1.2 billion last year — but the Online Safety Act still treats deepfakes as "content" not "crime." The police close files before they're opened.
The regulators are missing in action
The ICO has enforcement powers over automated decision-making. It has used them twice since 2018. The CMA talks a good game on foundation model competition while Microsoft embeds Copilot into every Office 365 licence — locking in every UK council and SME by default. Ofcom's draft codes for illegal content treat generative AI as a hosting problem, not a supply-chain one.
Meanwhile, the AI Safety Institute — £400 million of taxpayer money — publishes evals on frontier models that don't exist in the wild. They're testing GPT-5 capabilities while GPT-4o powers the chatbot denying your Universal Credit appeal. DWP's own equality impact assessment admits "disproportionate adverse effects" on disabled claimants. The Minister? Silent.
Small businesses carry the liability
Ask a Sheffield creative agency who's liable when Midjourney spits out Getty watermarks. Ask a Birmingham recruiter whose AI screening tool rejects every CV with "career break" — disproportionately women. The EU's AI Act makes the deployer responsible. The UK's "pro-innovation" framework makes the victim responsible.
Peter Kyle promised "binding requirements" for the most powerful models. The King's Speech delivered a voluntary code. DSIT calls it "agile." I call it capture.
The morning briefing is simple: the harms are here, the laws are gaps, and the people paid to protect you are briefing against regulation. Who exactly is minding the shop?