18 Aug 2026
Noggin HQ has won FCA authorisation as a credit reference agency, one of only a handful in a decade, and raised £2.3m to sell lenders open-banking data on what borrowers can actually afford. It could help the wrongly declined, but it means sharing your bank transactions.
18 Aug 2026
Sir Michael Caine has given his voice to University of York and Southampton researchers testing whether people can tell his real voice from an AI clone. The premise is uncomfortable: our ears are far worse at catching fake audio than our eyes are at catching fake images.
18 Aug 2026
Versatile RobotX, spun out of the University of Essex, has raised just over £1m, part of it public money from Innovate UK and Defra, to move its AI fruit-picking robots from trials to working farms. It is a concrete, small answer to a big question: what does AI for agriculture actually look like?
18 Aug 2026
The UK Atomic Energy Authority has spun out Singular Machines, a company selling "agentic AI" that helps engineers manage hugely complex projects while keeping humans in charge of the decisions. It is a rare, concrete look at agentic AI aimed at a serious, safety-critical job.
17 Aug 2026
In a US lawsuit over a deadly explosion, an expert hired by 3M used ChatGPT to write most of his "expert report", and the prompts, including one asking it to show the company was "0% at fault", came out in court. It is a sharp lesson in how AI is quietly entering expert evidence, and why UK rules on expert duties matter.
16 Aug 2026
OpenAI has switched on Computer History, a Recall-style tracker that logs what you do on your Mac and turns it into a timeline ChatGPT can search. It is opt-in, paid-only, and not launching in the UK, the EEA or Switzerland.
15 Aug 2026
The AI-evaluation group METR stopped trusting the press releases and plotted the actual numbers. Its finding is lopsided: a sharp acceleration in one area only, finding software security holes, with maths up modestly and algorithm design flat. And most of the security surge is in bugs that were disclosed, not ones attackers have exploited.
14 Aug 2026
British AI companies raised a record £7.2bn in six months, and 84% of it went to five firms in one city. Whitehall quietly bought workforce data, home-care sensors and a website for its 10-million-worker training promise. One security firm says it used off-the-shelf AI to turn a Zoom flaw into a working attack in under a day. Anthropic will hide a watermark in Claude's text, Spotify will badge AI artists, and the UK advertising regulator banned an "AI girlfriend" advert. All 20 stories we covered this week, grouped and in plain English.
14 Aug 2026
A three-week AI apprenticeship pilot for 16-to-21s has launched in Wigan, Blackburn and Blackpool, aimed at youth unemployment. It is small, just 70 places, but it is the test run for an England-wide scheme in 2027.
14 Aug 2026
A litigant buried invisible text in a US court filing, telling any AI that read it to rule in his favour. The judge caught it, sanctioned him, and warned the trick will spread. Here is what "prompt injection" means, and why it matters beyond one court.
13 Aug 2026
A Midlands startup has secured £619k from Innovate UK to bring physics-based AI to steel manufacturing, framed as a bet on 'sovereign' British AI. It is early-stage and unproven, but a clear example of AI aimed at heavy industry rather than chatbots.
13 Aug 2026
Mindgard, spun out of Lancaster University, has raised $30m (about £22m) to expand a platform that attacks AI systems on purpose and reports the flaws. It is the second UK AI-security raise this week, and a sign the field is becoming its own category.