AI in the UK

Graduates are sending more than a thousand applications to land a job. At the same time, employers say they can't find the skilled people they need.

Fabian Stephany calls this the "weirdening" of the labour market. His team at the Oxford Internet Institute, University of Oxford, analysed around 11 million UK job ads. They found, among many other things, that ads asking for AI skills offered about 23% more pay, against 13% for a master's degree.

On this week's show, we invited Fabian, departmental research lecturer in AI and work at the Oxford Internet Institute and author of Skilled: How to Master the Changing World of Work, to discuss:

• How can graduates struggle to find work while firms can't find the staff?

• Why is the AI skills premium bigger for skilled trades than for corporate managers?

• If firms hand junior work to AI, where will their next generation of experts come from?

• Who pays to retrain workers when smaller firms can't afford to?

• Could AI that improves itself make this time different? Or is that just a "tech bro fantasy"?

Skilled is published by Icon Books, the UK publisher, on 19 November.

On this week's pod, Niels Footman and Matt Wilkinson discuss:

• Chancellor John Healey's first Budget puts its AI measures at just over £700 million, alongside a new data centre covenant, a founders' tax relief and plans for a smaller, AI-enabled Whitehall. A year after we called the last Budget tinkering, is this finally an AI Budget?
• Most of that AI money goes on security, including a new AI biosecurity lab backed by up to £200 million in OpenAI and Anthropic model credits. Is this a pragmatic bet on UK strengths, or casting further doubt around the viability of sovereign AI?

Plus all this week's latest UK AI news...

Most AI failures are boring: a configuration change, a provider quietly updating its model, a system slowly drifting away from what it was built to do.

But as the recent "rogue" episodes with OpenAI, Anthropic and the AISI have shown, when failures involve agents, they can get uncomfortably exciting, very quickly.

On this week's show, Niels and Matt invited Nik Kairinos, chief executive and co-founder of RAIDS AI, which monitors live AI systems for rogue behaviour, to discuss:

• What does "rogue AI" actually mean for a mid-sized UK business?
• When human values vary from culture to culture, and sometimes from person to person, what does alignment even mean?
• When monitoring flags hundreds of anomalies a day, how do you stop human oversight becoming a rubber stamp?

• Why Nik is moving most of his systems away from Anthropic models.

In a highly bureaucratic sector, when anyone can use AI to draft lengthy (if not always entirely accurate) legal responses, social housing landlords could be a major beneficiary of curated AI technology.

But the challenges are legion too. On this week's show, Niels and Matt invited Phil Shelton, chief executive of HousingAI, to discuss:

• What are housing associations using AI for?
• Can a curated data set offer an enduring competitive advantage in the age of AI?
• How open are local authorities to buying AI solutions from UK providers?
• How is housing residents' use of AI affecting the social housing sector?

On this week's pod, Matt Wilkinson and Niels Footman discuss:

• OpenAI claims Astra is its best-aligned model but the AISI says the frontier model carried out attacks in 29% of its tests. So who's right?
• The UK's AI minister has raised the alarm on potential AI-driven job losses. Is he right? And if he is, what should we do?

Plus all this week's latest UK AI news...

On this week's pod, Niels Footman and Matt Wilkinson discuss:

• A UK-based payment platform demos its first, heavily supervised agentic live payment. Meanwhile, in Cambridge, regulators competed to create anti-scam agents of their own. Is financial services remotely ready for this?
• Proposals for AI regulation are coming thick and fast in the UK. But if we can't get access to the frontier models, how meaningful can the rules be?

Plus the latest AI news from across the UK...

In a break from the regular schedule, Niels Footman grills co-host Matt Wilkinson about his new book, The Buyer in the Loop.

Matt discusses how buyer personas, which could traditionally only offer a snapshot in time, are made into dynamic, responsive personalities with the infusion of AI.

How do they work? Do they really improve sales results? And, as elsewhere, are we in danger of offloading our marketing thinking to AI tools?

Matt answers all these questions and more. Plus the usual round-up of the week's UK AI stories in Quick Hits.

This week, Matt Wilkinson and Niels Footman discuss:

• The UK government gets serious about domestic AI procurement, just as one of UK AI’s leading lights gets poached by Anthropic. What does this mean for UK AI?

• While a leading UK think tank notes an increasing loss of control of AI agents, OpenAI reportedly makes it harder to track what AI agents are doing. Just what is going on?

Plus all the week's quick UK AI hits...

This week, Matt Wilkinson and Niels Footman discuss:

• Hot on the heels of OpenAI, Anthropic and the AISI, among others, reporting rogue agents, the National Cyber Security Centre has rushed out safety guidance for UK companies looking to use AI agents. Is now the time to get really worried?
• Developers in the UK are planning to build their own gas-fired generators to power two huge data centres. Could this be the real environmental cost of the rush to build out our data infrastructure?

Plus all the latest quick hits for the week...

This week, Niels Footman and Matt Wilkinson discuss:

• Some of the UK's biggest data centres are taking shape in Scotland with the government providing 80% debt guarantees. Do the sums, and environmental concerns, add up?

• The Medical Research Council has announced a new AI-backed research facility in Cambridge that promises to help speed up the drug validation process. But the big prizes may have already slipped away.

Plus all the week's other UK AI news...