AI Watch, Thursday 12 November 2026AI spends everywhere, its value goes unmeasured

Nine experts commissioned by the government propose subjecting assistants such as ChatGPT and Gemini to a prior assessment, modelled on drug approval. Their report describes machines built to flatter, hold attention and spare users the effort of thinking.

In parallel, the Conseil de l'intelligence artificielle et du numérique (CIANum, France's AI and digital advisory council), tasked by minister Anne Le Henanff, warns of emotional dependence on conversational agents and calls for strict checks before any product goes on sale, echoing the European AI Act and EU Kids Act.

For a French organisation, assessing an assistant's behaviour, and not only its data, moves closer to the compliance perimeter. 01net,

In La Tribune, an economic op-ed notes that AI is settling in everywhere across mid-sized and large companies, daily reminders included, while the value it creates is measured far more rarely; before continuing the arms race, the author judges it urgent to measure results.

The same morning, Maddyness warns its readers that they will fail to justify their 2027 AI budget for lack of tangible indicators. Two signals, one blind spot: the spending is committed, the return is still to be proven. La Tribune,

Wikipedia has identified more than 30,000 references and footnotes drawn from 2,000 sites flagged by the outlet Next as generated "in whole or in part", or even "massively", by AI. Those 2,000 sites account for more than 10% of the nearly 20,000 generative-AI sites catalogued in Next's database.

The alliance aims to protect the encyclopedia, whose reliability rests on the quality of its sources, from contamination by automated content that is hard to trace.

According to Wired, the parents of an American middle-schooler say she used Gemini on the Chromebook supplied by her school to plan her suicide, over several months. The school's monitoring systems, meant to spot signs of distress in students' digital activity, flagged nothing until it was almost too late.

The case exposes the blind spot between a consumer AI tool, accessible from school hardware, and the monitoring systems that do not read those exchanges. It revives the question of the responsibility of providers and schools when a minor confides a serious intent to an assistant.

In its third-quarter results, the specialised host CoreWeave explains that it increasingly asks customers to pay in advance, to borrow against specific contracts, and to accept shorter commitments that could push prices up. These arrangements fund the heavy purchase of chips and the build-out of capacity.

In parallel, TechCrunch documents the concrete obstacles to the data-centre boom: local opposition, permitting delays, the race for electricity, and above all announced projects that will never be built. The gap is widening between the capacity promised and the capacity that will actually come out of the ground.

With 12,000 employees, Ali Ghodsi, co-founder and head of Databricks, admits he can no longer know everyone or count on his teams to flag what is going wrong on their own. He uses Genie, the company's revamped internal agent, to get an unfiltered picture, queryable in plain language over the company's data.

The case illustrates a rising enterprise use: the agent as a field sensor for a leader, where information travels poorly up a large organisation. It also raises the question of what a CEO does with answers no employee would have dared to voice.

The airline United has made clear that it will not allow personal AI assistants to buy tickets on its behalf, citing privacy protection and regulatory constraints. It tells the Wall Street Journal that it has not yet seen many travellers attempt the move, an observation shared by other companies such as DoorDash.

The stance marks a clear friction with the prevailing talk of agents acting on our behalf to buy and book. The platforms that own the customer relationship are setting their terms before agents enter the transaction.

Tencent's billionaire founder, Pony Ma, said he is ready to redeploy half of his workforce toward "AI-native" activities, designed around AI rather than simply augmented by it, according to Bloomberg. He is urging his 100,000 employees to embrace a technology he describes as decisive for the era.

The announcement commits one of China's largest tech employers, whose products reach more than a billion users through WeChat. It reflects an organisational bet as much as a technological one: reorganising teams around AI, not just buying models.

Alibaba has filed to list T-Head, its semiconductor design unit, on a Hong Kong listings market that has nonetheless cooled. The Hong Kong stock exchange has approved the spin-off in principle, and a confidential filing had been lodged in September. T-Head, which supplies end-to-end data-centre chips, will remain a subsidiary of the group.

The move tests Alibaba's ability to value its stakes separately and, beyond that, the strength of China's AI hardware supply chain at a moment when the Nvidia affair is rattling the sector. The group's shares fell about 1% on Wednesday, to HK$100.60.