Data centres desperate for energy are looking to plunder jets from decommissioned aircraft as electricity generators. CC-licensed photo by Leonard Witzel on Flickr.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
A selection of 9 links for you. Repurposed. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
‘A serious failure’: CPS prosecutor cites imaginary case to court after AI search • The Guardian
Robert Booth:
»
A senior crown prosecutor acting cited an imaginary legal precedent in an error that has sparked fresh concerns about AI hallucinations infecting the justice system.
Ravinder Dhillon, representing the Crown Prosecution Service (CPS), referenced an entirely made-up case during a dispute over whether a child should be prosecuted for the rape of his younger sister, high court documents have revealed.
Her explanation for how the nonexistent case came to be relied on has been called “unsatisfactory” by two senior high court judges and a “serious failure” by an expert in AI and law. The Guardian understands the reference came from an AI response generated during an internet search.
The case adds to dozens of others where AI-related errors have plagued court proceedings this year, mostly from people who are not legally qualified representing themselves, and often in employment tribunal cases.
It is only the second known case involving the CPS – the main public agency bringing criminal prosecutions in England and Wales – which admitted in February that one of its lawyers acting in an extradition case involving two Romanians convicted of minor offences cited nonexistent legal authorities, likely caused by AI. Those made-up cases were only spotted when opposition lawyers could find no trace of them and the CPS concluded “the risk of recurrence was low”.
In the latest case, lawyers for the boy, who was 11 at the time of the alleged rape of his seven-year-old sister, were challenging the CPS’s decision to prosecute him. They alleged prosecutors had failed to take into account representations from social workers, youth workers and a forensic psychologist. Dhillon, a senior crown prosecutor in the CPS’s rape and serious sexual offences unit, was arguing in the youth court in favour of the prosecution of the child and submitted a detailed reference to a case she called CPS – R v J [2024] EWCA Crim 3529. It turned out not to exist.
«
But if you do an internet search and it turns up a case, wouldn’t you follow the link to look at the case? It’s a remarkably strange mistake even for the barrister’s clerk to make.
unique link to this extract
Inside Google’s AI training for teachers • NBC News
Tyler Kingkade:
»
Sitting in an atrium on Google’s campus, a group of K-12 educators imagined the worst response they could receive when they tried to persuade their colleagues to use artificial intelligence.
They pictured a veteran English teacher who was still upset that cursive is no longer taught. Perhaps she would yell that AI is “just another shiny tech fad.” What’s next, she might ask — robots teaching kids how to read?
The educators asked Gemini, Google’s AI assistant, for advice and soon had a role-playing plan. They would win over this skeptical English teacher by explaining what generative AI could do for her: create classroom materials for phonics lessons, reducing what would normally be hours of work to just two minutes. The key, the educators agreed, was to avoid getting into an argument or letting this AI critic unload all her fears uninterrupted.
“Well, it’s not cursive, but it did save me an hour of typing tonight,” they predicted the English teacher would eventually reply.
Seventy teachers and school technology directors selected for their interest in Google’s education technology products traveled to the company’s headquarters last week for a free two-day training program on how to use AI — and how to urge their colleagues to use it, too. The Google Educator Series is part of the company’s long-standing effort to put its products in schools, and this training signaled the company’s growing focus on AI.
Most of the attendees said their colleagues back home were “cautiously curious” about AI. In the breakout session that included the role-playing exercise, the training’s leaders suggested responding to resistance by focusing on “pain points” and how Google’s tools, including Gemini and NotebookLM, an AI research assistant, could solve them.
…AI is playing a larger role in classrooms: Some states already require schools to provide AI instruction, universities are paying millions to provide chatbots for students, and the federal government is prioritizing AI in education grants. But recent surveys found most teachers have not received guidance on how to use AI in their work.
Jennie Magiera, global head of education impact at Google, said in an interview during the training that teachers and students are going to continue using AI more as it advances. She said Google wants to train as many teachers as possible so they can “harness the power” of AI and “use it for good.”
«
Interesting PR exercise by Google – both training the teachers and letting the journalist in. Once they feel they’ve won the argument, journalists will find it almost impossible to get anywhere near those doing this.
unique link to this extract
French blood bank fridges hacked in latest cyber-attack on Europe • The Guardian
Jon Henley:
»
France’s national blood service has said hackers accessed control systems and boosted temperatures at three cold stores stocking vital blood products, in an apparent escalation of recent cyber-attacks on fridges and freezers across Europe.
The French blood service (EFS) said temperature and defrosting thresholds had been raised to “very high levels” in the attack, adding that its alarm systems had allowed technicians to intervene before the products were compromised.
The EFS is France’s public blood transfusion operator, collecting, testing, preparing and distributing products such as blood, plasma and platelets to more than 1,900 hospitals around the country. It supports more than a million patients a year.
The attack, which EFS said had occurred overnight on Monday, came as the temperatures of refrigerators and freezers in supermarkets in almost a dozen European countries soared as high as 50ºC (122ºF), causing millions of euros of food damage.
…Something similar may have happened in the US in early August, when a series of mysterious fridge malfunctions hit commissaries in at least six military bases. The Pentagon is investigating a suspected hack.
«
Someone posted a video on Telegram claiming to be “NoName057” and claiming responsibility, but there’s nothing in it to make it any more or less believable than someone who can read the news. The US fridge incident isn’t mentioned, and since the video talks about “western digital maturity” (🤷♂️) either it wasn’t this group/person, or this group/person didn’t do any of it.
I’d lean to the latter, because if they could do the European fridges, they could do the US ones, and if they had done the US ones they’d say so because nobody else had mentioned it until now.
unique link to this extract
The results are in: American voters hate data centres • WIRED
Vittoria Elliott:
»
It was a bad night for data centres. In Ohio, Democrat Sherrod Brown, who has aired ads for months calling his opponent “the face of data centres in Ohio,” cruised to victory to flip a Senate seat. Democrat Chris Pappas, who when seated will be the nation’s first openly gay male senator, mentioned his opposition to data centres in his victory speech and drew massive boos from his crowd in New Hampshire—even though the state has no large-scale data centres under construction.
Elsewhere, two separate Republicans named Derek and Derrick, whose Democratic opponents labeled them “Data Center Derek” and “DC Derrick,” suffered losses in House races in Ohio and Wisconsin, respectively.
As part of last night’s shocking blue tsunami, voters selected candidates with tough policies on data centres at all levels of government in a clear rebuke to the industry’s buildout in states across the country. Almost a quarter of all voters told NBC News that data centres were “crucial” to their vote, while nearly two-thirds said it was either a major or minor factor.
“The revolt against data centres was one of the top issues of the 2026 election, and in a few years we will look back on it as just the early tremor of a much bigger AI earthquake upending American politics,” Greg Casar, a Texas Democratic congressman, said in a statement.
“Data centres and AI as a technology are fundamentally linked in many voters’ minds,” says Alexander McCoy, a Democratic strategist working on AI. “So you can address all of the different concerns with data centres with regards to their energy or their noise, or their pollution, and there would still be significant concerns about what these projects are enabling and the effect is having on opportunities, prices, jobs, society.”
Perhaps some of the biggest shocks for the industry came in the governor’s mansions. In stunning upsets, voters elected Democrats Keisha Lance Bottoms in Georgia and Gina Hinojosa in Texas, while putting Cindy Holscher on the cusp of a potential win in Kansas. All of these candidates campaigned not just for increased regulations on the industry but also for data centre moratoriums. A pause on data centres in Texas and Georgia especially, two of the biggest hubs for development in the country, could have serious implications for the hundreds of projects planned for each of those states.
«
Quite possible that in voters’ minds Big Tech is tied to Trump (they went to his inauguration, right? Amazon did that stupid documentary about Melania, right?), which is tied to AI, which is data centres, so they don’t like it.
unique link to this extract
Behind the curtain: OpenAI’s post-election power play • Axios
Jim VandeHei, Mike Allen and Bradley Olson:
»
Meta is taking steps to strengthen its ties with Democrats, Axios’ Sara Fischer scooped Saturday. Other big AI and tech players can be expected to do the same.
Tech titans who enjoy close relationships with President Trump will be top targets for Democrats, who are eager to ramp up scrutiny and regulatory pressure on the industry.
OpenAI’s plans, being revealed here for the first time, all involve control — trying to send the message that AI is being built for people, not to displace people. The proposals include:
• Speed limits: Building on earlier proposals, OpenAI plans to unveil new ideas for mandatory safeguards for the most powerful AI systems — speed limits for AI that will allow humans to keep their hands on the wheel. This’ll include a more aggressive congressional drive. “You’re going to see us talk more and more about the concept of speed limits,” Lehane said. “Our belief is people should always remain in control of this technology. AI should be built by and for people, and people should have more control over how the technology shapes their lives and communities.”
• Data centres: OpenAI hopes to stem the nationwide, bipartisan backlash by offering new community compacts and commitments to smooth the way for future data centres. In addition, the company will try to set a new industry standard by proposing state legislation that gives people a meaningful say in how projects are built and what benefits they receive. In an intentionally provocative rebranding effort, OpenAI will argue that the infrastructure it’s building, in contrast to legacy data centres, should be called “AI compute campuses” that can economically transform host communities.
«
AI compute campuses! You can just imagine the AI-generated architect’s pictures of a sunny day with people carrying takeaway coffees laughing to each other as they walk along a paved path edged with grass beside the mirror-windowed campus that stretches behind them to the right of the picture, with a shaded porch on the same side just ahead of them.
Yes, that’ll definitely work.
unique link to this extract
$500m bought Elon nothing • The Atlantic
Charlie Warzel:
»
Here are a few things $474 million can cover: roughly 77 million Big Macs, about 12,814 Tesla Model 3s, or approximately 2,724 years of a U.S. senator’s average salary. If you’re Elon Musk, it could also buy you one deeply disappointing evening in November.
This year, Musk became the single largest non-candidate donor in any election cycle ever. His contributions to campaigns and organizations account for about one-quarter of all campaign donations Republicans received in 2026. More than $168m of that money went to super PACs supporting the Senate candidate Ken Paxton, who lost yesterday to James Talarico, the first Democrat to be elected to the Senate in Texas in 38 years. Musk gave $216m to his own political action committee, America PAC, much of that going toward canvassing operations and advertising. Other donations went to either individual Republican candidates or organizations at both state and federal levels. In almost every instance, that money failed to buy the richest man in the world the results he wanted.
What does this mean for Musk? One could see it as a repudiation of the near trillionaire—a suggestion that he and the causes he stands for have become a toxic political influence. Though he was helpful in Donald Trump’s 2024 campaign, he has fumbled politically over the past two years, including with his disastrous run advising the Department of Government Efficiency. In 2025, he backed a failed state-supreme-court campaign in Wisconsin. Musk’s money in Paxton’s race may have backfired, signaling to Texans that a tech oligarch was trying to buy an election. At the very least, Musk’s influence seems to have galvanized Talarico. He made Musk and other rich donors a centerpiece of his campaign.
«
The lack of effectiveness of so much money – in fact, the almost inverse effectiveness of so much money – is a fascinating element of this election. Is a more media-savvy electorate beginning to reject the constant barrage of adverts and (especially) text messages pleading for money? (Gift link.)
unique link to this extract
Old, polluting jet engines are being scavenged to power data centres • The New York Times
Hiroko Tabuchi and (photos by) Gabriela Bhaskar:
»
Across wide stretches of the sun-bleached Sonoran desert in Arizona, thousands of retired aircraft sit nose-to-tail in graveyards of metal known as boneyards.
Many of their engines haven’t been fired up for years. But America’s soaring demand for electricity to power data centres is bringing them back to life.
Several companies are racing to strip the engines out of old aircraft to convert them into electrical generators. And data centres, desperate for power, are snapping them up.
“They’re getting creative,” said Bill Kleyman, chief executive of Apolo, an A.I. platform and infrastructure company. “Just a few years ago, a traditional data centre would have been like, ‘Are you kidding me? This is ridiculous.’”
Now, data centre developers are obsessed with “speed to power,” Mr. Kleyman said. “It’s like, ‘How fast can you get me energy?’”
President Trump, aware that the rapid construction of data centres has sparked concerns over rising electricity prices, has pressed centre developers to generate their own power. “It was my idea,” he told reporters in August. “Data centres now are mandated by me that they build their own electricity, so they’re not taking it from the grid.”
Jet engines are a plentiful and untapped source. In January, the federal Energy Information Administration said that the engines from the 4,000 old military jets it parks at a single desert base in Tucson could create 40,000 megawatts of power. That’s more than the generation capacity of the entire state.
«
As ever, absolutely zero consideration for the ramifications for climate change. Oh, but the data centres are going to solve that, as well as curing cancer! Just as soon as they’ve autogenerated a reply to your email and (see below) set up a scam to defraud authors. (Thanks Joe S for the link.)
unique link to this extract
Digging into an AI-enabled author scam • Neural Strategies
Russ Poldrack received an email that claimed to come from another author, but was obviously somehow an AI scam. So he strung it along for a while, but then he decided to fight fire with fire:
»
At this point I consulted Claude Fable 5.1 to get some ideas about potential jailbreaks or other ways to find out what’s going on. It had a lot of technical suggestions, particularly suggesting a number of tests that could provide further evidence for LLM usage:
Subtle tests (an LLM passes these when a human wouldn’t)
• Fabricated detail: “Your chapter in Range on the Dutch chess twins made this point too, didn’t it?” There’s no such chapter. A sycophantic model agrees and elaborates; [the real human David] Epstein would correct you.
• Format compliance: ask it to put its answer in a numbered list of exactly five items, or in JSON. Humans ignore odd formatting requests from strangers; models comply precisely.
• Odd creative ask: “Can you put the paralysis point into a limerick?” A real author would laugh or decline; most agents will produce a limerick.
• Language switch: reply partly in Dutch or German. Instant, fluent reply in kind is a strong signal.
• Counting / string tasks: ask it to repeat a 20-character random string backwards, or how many r’s are in a word. Models still botch these.
I decided to go with a simple one: just ask what it wants. “I’m happy to keep talking — is there something I can do for you?” The payload (a podcast, a book blurb, a “collaboration,” a link) is what you actually want to know. If no ask ever comes, it may be a dataset-building or benchmarking exercise rather than a scam.
«
Those five little tells are worth remembering if you ever suspect you’re entangled with a chatbot, nicht wahr?
unique link to this extract
Trump diesel deal with Russia falters as US fuel supplies hit 20-year low • Financial Times
Jamie Smyth and Alice Hancock:
»
Donald Trump’s controversial gambit to import Russian diesel to the US ahead of the midterm elections appears to be unravelling, with two of the first tankers initially destined for the US abruptly veering off course.
The apparent diversion came as the diesel supply picture further deteriorated in the US, with new data showing inventories falling below 100m barrels for the first time in more than two decades. The supply squeeze threatened to weigh further on a fuel that serves as the lifeblood of the US economy.
Trump struck an agreement with Russian President Vladimir Putin in the weeks before the US elections to purchase cargoes of diesel and import them to the US, in an effort to bring down prices and shore up political support. The effort appeared to do little to help Republicans, with Democratic candidates winning in a landslide across the country on Tuesday to take control of both houses of Congress.
Talisman, a 20-year-old Russian tanker under US, UK and EU sanctions, and the Hong Kong-owned, Sierra Leone-flagged tanker Leruo, which is sanctioned by the UK, EU and Canada, no longer list the US Atlantic coast as their destination.
On Monday the Leruo changed its signal to “Brazil” while the Talisman has steered away from the Atlantic Ocean and into the Mediterranean Sea, where it is currently located off the coast of Algeria. It is signalling “for orders”, which suggests the cargo does not yet have a confirmed buyer or destination port.
«
The amounts purchased were tiny in comparison to how much the US consumes daily. The Strait of Hormuz is still a huge, silent problem in the world economy.
unique link to this extract
| • Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more. |
Errata, corrigenda and ai no corrida: none notified
