AI Watch, Thursday 5 November 2026Data centres, the midterms' big losers

The European Data Protection Board, the body that brings together Europe's national data protection authorities, has opened "Privacy for Kids" together with France's CNIL (the country's data protection authority), a platform devoted to digital literacy for the youngest users.

The initiative accompanies a wider regulatory push on the protection of minors online: an independent study commissioned by the European Commission finds that one child aged 9 to 11 in three, across six European countries, reports holding a social media profile.

Anthropic has unveiled FlashProve, a tool that proves a response genuinely comes from the advertised model, rather than a degraded or substituted version behind the scenes, Clubic and 01net report.

The need is concrete: when a company calls a model through a programming interface (an API), nothing today guarantees it is not being served a cheaper model under the promised label. Anthropic has, however, filed the tool under research, with no commitment to deploy it.

Hackers resell access to the most expensive AI models on the cheap by hijacking corporate credentials, leaving the companies with the bill, sometimes up to a million dollars, Clubic reports. The method consists of stealing an organisation's access keys to an AI service, then selling that access to third parties for as long as the fraud goes unnoticed. The practice is growing and targets poorly protected business accounts.

La Tribune visited the centre in Erskine, Scotland, where Hewlett Packard Enterprise tests, repairs and returns to the market Nvidia graphics cards, the processors coveted since the AI rush. Two hundred staff recover components that large data centre operators are shedding.

The site sheds light on a rarely seen side of the AI economy: a secondary market for hardware is taking shape as first-generation chips give way.

At its earnings presentation on Wednesday, Uber revealed that third-party AI agents had arranged only four trips during its busiest week of the third quarter of 2026, The Verge and TechCrunch report. On the delivery side, Uber Eats counted "around 500" orders placed by agents over the same period.

For a platform that handles tens of millions of rides, these volumes are close to nothing. Executives cited the figure to reassure on the risk of disintermediation, the fear of seeing AI assistants slip between Uber and its customers. The observation goes beyond Uber: autonomous booking by an agent, heavily announced, remains a marginal practice.

The British government had claimed that AI would add £400 billion to its economy, a figure drawn mostly from an estimate supplied by Google, Wired reports. The minister responsible for AI now concedes that the realistic amount is under a third of that sum, on the order of £120 billion. The researchers cited point to slow adoption by companies as the main cause of the gap.

The episode shows how easily an impact figure, taken up by the authorities, swells well beyond what real usage justifies.

Alphabet saw queries on its engine grow at their fastest pace in five years, but search advertising revenue growth slowed to 12% year on year in the third quarter, against roughly 17% the previous quarter, The Verge reports. AI-generated answers, the AI Overviews, sit at the top of the pages and divert some of the clicks that fed the ads.

Google is now inserting advertising into those answers, without so far recovering the profitability of the classic blue link. The case raises a question common to every service provider: AI is costly to produce and, for now, earns less than what it replaces.

Isomorphic Labs, the drug-discovery company spun out of Google DeepMind, is reportedly in talks to raise about $5 billion, Bloomberg reports. Led by Demis Hassabis, it applies to treatment design the methods that produced AlphaFold, the system that predicts the structure of proteins. The funding round, if confirmed, would rank among the largest in AI-applied healthcare.

It shows that capital keeps flowing toward scientific uses, even as other parts of AI see their promises revised downward.

At a panel of the World Internet Conference in Wuzhen on Tuesday, engineer Gao Xi reaffirmed DeepSeek's commitment to open source to speed the arrival of artificial general intelligence, an AI as versatile as a human, even if that openness increases safety risks, the South China Morning Post reports.

He argues that open development fosters shared innovation and technical progress. The stance stands up to concerns voiced in both the United States and China, where the authorities are questioning the threat such a strategy would pose to national security. The Chinese lab is owning a line that runs against the caution displayed by several major players.

China has barred the sale to SoftBank of Swiss group ABB's industrial robotics unit, valued at $5.4 billion, Bloomberg reports. The veto thwarts the Japanese group's ambitions in physical AI, that is, robots equipped with AI capabilities.

It illustrates Beijing's use of its merger review as leverage in the technological rivalry, mirroring the restrictions Washington imposes on semiconductors.