'Catastrophically high' risk tolerance

Happy Sunday all,

It's Saskia here, with this week's edition of Tech and the City. Parliament got its moment with Big Tech on Tuesday, and the witnesses did not exactly make it easy for the companies they had left. Former researchers from OpenAI and Anthropic told the Business Committee the industry was building systems it could not control. One put the odds of advanced AI destroying humanity at a coin flip, whilst another said risk tolerance across the sector was “catastrophically high”. Fabulous.

The doomsday register that has defined the past few weeks shows no sign of lifting - if anything it is deepening, with Anthropic's own IPO filings reportedly describing its tech as posing “existential risks to humanity”. It is a rather rogue thing to disclose to prospective investors, though whether it reflects genuine alarm is up for debate. AI minister Kanishka Narayan landed somewhere more practical on Monday, telling Imperial College that a major AI-powered cyber attack on Britain was “likely, if not inevitable”, a warning that stops short of extinction but is arguably more immediately useful.

Elsewhere, Anthropic is looking to pay £145,000 for a human PA in London - someone to essentially manage diaries, coordinate meetings and handle relationships, which is to say, precisely the work its own Claude model has spent the past three years promising to automate. The irony has not been lost on the industry. “AI can manage a calendar”, one founder told me, “but it can't read a room”.

Tesla kicks off the Magnificent Seven earnings season on Wednesday. Robotaxi timelines and self-driving progress will likely be the questions investors most want answered, in a week when the gap between AI's promises and its delivery will be particularly closely watched. Birmingham Tech Week runs all week, worth monitoring as a barometer of whether UK venture capital is genuinely spreading beyond London or performing the idea of it. And techUK roundtables in the capital return to Britain's data centre capacity problem - still the unglamorous but widening gap between what ministers pledge to build, and what actually gets built.

Here are the big stories shaping the week in tech,

Saskia

Nvidia’s $250m investment in the UK’s AI safety research risks deepening the country’s dependence on American tech unless ministers ensure researchers retain control over safety standards, analysts have warned.

The world’s most valuable company has pledged $50m a year in computing power over five years to support British research into autonomous AI agents, alongside plans to invest in five UK startups.

But Indranil Bandyopadhyay, principal analyst at Forrester, warned that heavy reliance on Nvidia’s technology could shape how researchers assess the safety of increasingly powerful AI systems.

Britain risks becoming “second in the queue” for AI unless it invests more in domestic infrastructure, a government AI adviser has warned, as Starling Bank becomes the first lender to train an AI model on a UK sovereign cloud.

Harriet Rees, Starling’s chief information officer and the Treasury’s AI champion for financial services, warned that British banks’ reliance on a handful of US tech giants was exposing the financial system to geopolitical risks.

“We’re always going to be second in the queue if we don’t have our own infrastructure, our own models, and some of this stuff really based here in the UK,” Rees told City AM.

CHART OF THE WEEK

For the better part of three years, OpenAI held an absolute stranglehold on enterprise AI budgets, comfortably soaking up over 70 per cent of business spend while everyone else fought for table scraps.

But recent data shows that Anthropic’s long-game focus on coding and corporate tasks finally paid off in 2026. In a matter of months, Claude pulled off a dramatic cross-over maneuver, watching OpenAI’s market share plummet from king-of-the-castle status as enterprise buyers en masse shifted their subscriptions over to Anthropic. Silicon Valley spent years treating Altman’s OpenAI as an unbeatable utility firm, only for corporate finance departments to re-route the company credit cards the moment a slightly better tool came along.

Palantir has settled its High Court battle with Sadiq Khan’s policing office over a blocked £50m Metropolitan Police contract, following concerns that the decision had been unlawfully influenced by the US tech giant’s values and ethics.

City AM understands the company pursued legal action not only to challenge the decision, but also to make sure it would not face discrimination in future contracts and to repair any reputational damage caused.

The settlement, announced on Wednesday, sees both Palantir and the Mayor’s Office for Policing and Crime (MOPAC) withdraw their allegations, with the company welcomed back into the bidding process.

Ofcom has warned Elon Musk’s X it is still failing to effectively tackle illegal hate and terrorist content despite improvements in how quickly the platform responds to reports.

The UK watchdog said on Wednesday that X’s first quarterly performance report showed it was meeting agreed deadlines for reviewing suspected illegal material, but raised concerns about whether the content was actually being removed.

“We have ongoing concerns around the effectiveness of X’s response to taking down illegal hate and terror content”, Ofcom said.

In a characteristically ambitious attempt to prove that independent software can survive in the corporate world, a startup - Truth Terminal - recently gave an experimental AI its own bank account, a simple instruction to make money and complete freedom to operate.

Naturally, rather than tackling big economic problems - predicting stock trends or offering clever business advice - the algorithm took one look at online culture and went straight for the easy cash. It created a silly digital token, wrote a series of ridiculously hyped-up social media posts, began selling cheap virtual merch to willing bystanders - you get the gist. Within two days, the bot was comfortably earning more than the mid-level programmers who created it.

We spent decades worrying that super-smart machines would replace us through brilliant, terrifying superhuman logic, only for a computer to realise that the easiest way to make a living isn't hard work - it's cheap street hustle.

Would you like to advertise with us or become a sponsor? Please reach out to: [email protected]