AI Daily Brief: 25 May 2026

25 May 2026

Three UK stories frame today's briefing: a mayor versus the Met over a £50m AI contract, Scotland's datacentre greenwashing problem, and UK businesses playing fast and loose with the word 'AI'. Globally, OpenAI is heading for a $1 trillion stock market debut.

Palantir accuses Sadiq Khan of 'putting politics above public safety' after £50m Met contract blocked

London Mayor Sadiq Khan has blocked a proposed £50 million contract between the Metropolitan Police and US data analytics firm Palantir, citing a 'clear and serious breach' of procurement rules. The deal would have seen Palantir automate aspects of criminal intelligence processing for Scotland Yard.

Palantir's UK and Europe head Louis Mosley responded sharply, accusing Khan of politicising procurement decisions. The company also works for the Israeli military and the Trump administration's immigration enforcement operations - links Khan has previously said conflict with the 'values of our city'.

The Met described the decision as 'disappointing', warning that without new technology it would need to cut officer numbers. The dispute exposes wider tensions inside Labour: the UK government holds a £330m NHS England deal with Palantir and a separate £240m Ministry of Defence contract.

Our take: The Palantir row is not just a London procurement dispute - it is an early test case for how UK public bodies will handle AI vendor selection when those vendors carry political baggage. Khan's decision creates a precedent where the ethics and geopolitical associations of an AI supplier become grounds for rejection, separate from capability or price. For any UK organisation procuring AI tools, this signals that governance frameworks need to cover vendor values and associations, not just technical compliance.

Scotland's 'green datacentre' policy has no definition of green - and AI is about to expose that

Scotland is actively courting major AI datacentre investment, with an AI growth zone near Glasgow claiming £8.2 billion in private backing and more than a dozen facilities currently seeking planning permission. Collectively, these would consume roughly 6.2GW of power - one and a half times Scotland's peak winter demand for the entire country.

The problem, according to Action to Protect Rural Scotland, is that the country's 'green datacentres' policy was written in 2022, before ChatGPT launched and before AI-driven power demand became a known factor. There is no working definition of what qualifies as a green datacentre, meaning AI facilities could claim the label while their full emissions impact goes unaccounted.

Green MSP Ariane Burgess called for urgent transparency: 'So far, the answers we've been getting out of the Scottish government have not provided any clarity.' The UK's National Energy System Operator CEO Fintan Slye has separately been encouraging datacentre developers to build in Scotland for its renewable energy advantage - though critics note that renewable connection does not automatically offset consumption at this scale.

Our take: This is a policy gap that could become a credibility crisis. Scotland's pitch to AI investors rests heavily on its green credentials - but if the definition of 'green' was frozen in 2022, before anyone understood how much power AI inference and training actually requires, the entire framework is built on an outdated assumption. UK businesses evaluating datacentre providers in Scotland should press for actual consumption data, not just a green label.

OpenAI files confidentially for a $1 trillion IPO targeting September 2026

OpenAI has filed confidentially with the US Securities and Exchange Commission for an initial public offering, working with Goldman Sachs and Morgan Stanley toward a September 2026 listing. The company is currently valued at approximately $852 billion by private investors, and analysts expect the IPO to push it past the $1 trillion mark.

The filing comes as Anthropic separately disclosed that it projects $10.9 billion in Q2 2026 revenue - more than double Q1's $4.8 billion - with its first-ever operating profit of $559 million expected for the period. The two companies are taking divergent paths: Anthropic arriving at market with profitable quarters already in hand, while OpenAI remains pre-profitability and is asking public investors to fund its path to positive cash flow, currently projected around 2029 to 2030.

A Forbes analysis notes that OpenAI's cumulative losses before profitability will likely run into hundreds of billions - far exceeding the $3 billion Amazon accumulated before its first annual profit in 2003, which AI bulls frequently cite as a precedent for patient investing.

Our take: The OpenAI IPO will be the defining moment for how public markets price the AI industry. If it lists at $1 trillion on negative cash flow, it will set expectations - and valuations - across the sector that UK AI companies and their investors will feel for years. Anthropic's profitable quarter, by contrast, shows that AI revenue is real and growing. The question is which model becomes the template: disciplined enterprise revenue growth, or blitzscaling toward platform dominance.

UK companies are demanding PR firms call their basic automation 'AI' - and the industry is pushing back

A Guardian investigation has lifted the lid on a growing practice among UK businesses: pressuring communications agencies to pitch ordinary automation tools, legacy software, and rule-based systems as artificial intelligence to capitalise on investor and media interest in the technology.

PR executives describe being asked to reframe products with no generative AI component as 'AI-powered', with one publicist noting: 'You can almost hear the eyes roll when you mention the word AI to a reporter.' Imran Ariff of London agency Fight or Flight says brands are 'drinking their own Kool-Aid' when overstating AI capabilities. Examples cited include a shoe company pivoting to acquire AI graphics processing units, and press releases about 'AI-powered basketball hoops'.

The practice mirrors greenwashing in environmental claims, with critics arguing that AI washing misleads buyers, erodes trust in genuine AI products, and creates regulatory risk as the FCA and CMA increase scrutiny of technology marketing claims.

Our take: AI washing is the reputational risk that UK business leaders are not taking seriously enough. When every piece of automation gets called AI, genuine capability becomes harder to distinguish, harder to sell, and harder to buy. For businesses evaluating AI vendors right now, the implication is clear: ask suppliers to be specific. What model? Trained on what data? Doing what, exactly? Vague claims of 'AI-powered' should be a red flag, not a selling point.

OpenAI's reasoning model solves an 80-year-old geometry problem - without being asked to

OpenAI has announced that one of its general-purpose reasoning models has produced an original mathematical proof disproving a conjecture in discrete geometry first posed by Hungarian mathematician Paul Erdos in 1946 - the planar unit distance problem. The company calls it 'the first time AI has autonomously solved a prominent open problem central to a field of mathematics'.

The proof was not produced by a system designed for mathematics. It came from a general reasoning model, suggesting that advanced reasoning capabilities are now transferring to domains the model was not purpose-built for. The result has been validated by prominent external mathematicians including Noga Alon and Melanie Wood, and Thomas Bloom who maintains the Erdos Problems website.

OpenAI was careful to note this corrects a previous embarrassment: seven months ago, a premature claim that GPT-5 had solved ten Erdos problems turned out to be wrong - the model had found known solutions in the literature. This time, independent verification came before the announcement.

Our take: The significance here is not the geometry problem itself - it is what general-purpose reasoning transfer implies. If a model trained as a general reasoner can autonomously solve an open research problem in a specialist domain it was not tuned for, the question for every UK business is: what does that mean for the knowledge work assumed to be safe from automation? Scientific research, legal reasoning, complex analysis - the boundary between 'AI tools' and 'AI colleagues' is moving faster than most organisations have planned for.

Quick Hits

• Anthropic disclosed it expects $10.9 billion in Q2 2026 revenue and its first-ever operating profit of $559 million, more than doubling Q1 figures.
• The UK AI Safety Institute, staffed by OpenAI and Google alumni, is being held up as an international model for countries developing AI oversight frameworks.
• Spotify and Universal Music have agreed a deal allowing subscribers to create AI remixes of licensed tracks, the first major label deal of its kind.

Frequently Asked Questions

How often is the AI Daily Brief published?

Every morning at 7:30am UK time, covering the previous 24 hours of AI news from over 30 sources.

How are stories selected?

UK-relevant stories are prioritised first, then by business impact and practical implications for UK organisations adopting AI.

Why should business leaders follow AI news?

AI is moving faster than any technology in history. Staying informed is essential for making smart decisions about AI investment, adoption, and governance.