NEW YORK

Donald Trump is arraigned in a Manhattan courtroom, the first former president ever charged with a crime

Thirty-four felony counts, a not-guilty plea, and a motorcade watched by the world. The republic did the ordinary thing and it felt extraordinary.

Donald Trump surrendered to the Manhattan district attorney on the afternoon of April 4, was fingerprinted, and sat at a defence table while a judge read 34 counts of falsifying business records against him. He pleaded not guilty. The charges, brought by District Attorney Alvin Bragg, concern a $130,000 payment made to the adult-film actress Stormy Daniels in the closing days of the 2016 campaign, and the way the reimbursement to Trump's then-lawyer Michael Cohen was booked as legal expenses. Each count is a felony; together they carry, in theory, years in prison, though a first offence of this kind rarely does.

It was the first time in the country's history that a former president had been criminally charged, and the machinery of it was deliberately, almost defiantly, mundane. There was no perp walk. Trump flew up from Florida, spent a night at Trump Tower, rode down to the courthouse in a motorcade broadcast live by every network, spent about an hour inside, and flew back to Mar-a-Lago, where he gave a speech that night attacking the judge, the prosecutor and their families. Legal scholars are divided on the strength of the case; the theory that elevates the bookkeeping misdemeanour to a felony, that it was done to conceal a second crime, is novel and will be fought over for months. A trial, if it happens, is unlikely before next year, in the middle of the campaign in which Trump is the Republican front-runner.

Three more investigations sit behind this one: the classified documents at Mar-a-Lago, the effort to overturn Georgia's 2020 result, and the federal January 6 inquiry, any of which legal observers consider graver than the case that got there first. Trump's campaign said it raised more than $10 million in the days around the arraignment.

TECHNOLOGY

An open letter asks the industry to pause; Italy pulls the plug on ChatGPT

Four months after ChatGPT, the backlash has a petition and a regulator. Neither is likely to slow anything down.

An open letter published by the Future of Life Institute on March 29 called on all AI labs to "immediately pause for at least six months the training of AI systems more powerful than GPT-4," warning of "profound risks to society and humanity." It gathered thousands of signatures, including Elon Musk, Steve Wozniak and a number of serious researchers, and also, in the first rushed days, a scattering of fake ones. Its critics split into two camps that rarely agree: those who said it overstated the danger, and those who said a six-month pause was both unenforceable and beside the point, since the labs racing hardest would simply not comply.

Two days later, on March 31, Italy's data-protection authority did something with actual force behind it: it ordered OpenAI to stop processing Italians' data and effectively blocked ChatGPT in the country, citing the lack of a legal basis for training on personal information and the absence of age controls. OpenAI took the service down in Italy while it negotiated. Regulators in France, Germany, Ireland and Canada said they were watching. The contrast was the point of the fortnight: the industry's own attempt to slow itself down was a letter with no teeth, and the only thing that actually stopped a product was a privacy law written years before anyone had heard of a large language model.

DEFENSE

Classified US documents appear on the open internet

A trove of highly classified American intelligence documents, including assessments of Ukraine's air defences and evidence of US spying on allies, surfaced on gaming and messaging platforms and then spread across social media in the first week of April. The Pentagon called it a serious breach and opened an investigation; officials spent the week telephoning allies to explain. The material appeared genuine, if in places crudely altered, and painted a bleaker picture of Ukraine's coming offensive than public statements had. How it got from a secure vault to a Discord server was, at press time, unknown.

EUROPE

Finland joins NATO

Finland became the 31st member of NATO on April 4, doubling the alliance's land border with Russia and ending three-quarters of a century of neutrality that the invasion of Ukraine had made untenable in a year. Its foreign minister handed over the accession documents in Brussels; the blue-and-white flag was raised at alliance headquarters. Sweden, which applied at the same time, remains blocked by Turkey and Hungary. Moscow said it would strengthen its forces in the west.

AMERICA

A school shooting in Nashville

A former student walked into the Covenant School, a small Christian primary in Nashville, on the morning of March 27 and killed three nine-year-old children and three adults with two rifles and a handgun bought legally, before police shot her dead. It was the deadliest American school shooting of the year so far and prompted the now-standard sequence: vigils, a fight in the Tennessee legislature that ended with the expulsion of two Democratic lawmakers who protested for gun control, and no change to the law.

IN BRIEF

Bakhmut; Ozempic; Twitter labels; a jury in California

The battle for Bakhmut ground into its ninth month, the longest and bloodiest of the war, with Russia's Wagner mercenaries claiming the centre of a city now almost entirely rubble. Demand for Ozempic and Wegovy, diabetes drugs that produce dramatic weight loss, outran supply and began reshaping how the food and pharmaceutical industries think about appetite. Twitter labelled National Public Radio "state-affiliated media," the tag it uses for propaganda outlets, then changed it to "government-funded"; NPR quit the platform. A California jury cleared Gwyneth Paltrow of fault in a 2016 ski collision on March 30 and awarded her the symbolic one dollar she had sought.

The Column

The six-month pause that was never going to happen

I deploy this technology for a living, in a small way. The letter asking everyone to stop misunderstands who "everyone" is.

By Daniel Reeve

I want to take the pause letter seriously, because some of the people who signed it are not cranks, and because the thing they are worried about, systems that get more capable faster than anyone can check them, is the thing I worried about in this space two months ago over a different story. But I have now read it several times, from the seat of someone who actually puts these models into other people's businesses, and I think it asks the wrong entity to do the wrong thing.

Here is the problem with a pause. "The industry" is not a body that can decide anything. It is OpenAI and Google and Anthropic and Meta and a dozen well-funded labs and ten thousand companies like the ones I work with, all making independent decisions under competition. A pause only works if all of them stop, and the only ones who would stop are the ones who were being careful anyway. The labs racing hardest have the most to lose from six months of stillness and the least reason to trust that their rivals are also standing still. You cannot disarm by asking nicely, one player at a time, in public, while the scoreboard keeps running.

What actually stopped a product this fortnight was not the letter. It was Italy. A privacy regulator, using a law written in 2018 for a world that did not have chatbots, told OpenAI it had no legal basis to hoover up Italians' personal data to train on, and no way to keep children out, and the product went dark in a country of sixty million people the next day. That is what force looks like. It is boring, it is specific, and it is aimed at a thing a company can actually be made to do: not "be less capable," but "show me your legal basis for processing this person's data."

This is the distinction I keep trying to make to clients who ask whether they should "wait until the AI is safe." Safe is not a milestone the technology reaches. There is no morning when the model stops being able to make things up. What there is, instead, is a set of concrete controls you can put around it: what data it may touch, who checks its output before it reaches a customer, what it is allowed to decide on its own and what it must escalate. Those are the same controls I would put around a new employee or a new integration. They do not require the vendor to pause. They require me to do my job.

So when someone asks me what I make of the letter, I say this. The instinct is right and the mechanism is fantasy. Nobody is going to pause. The people who could be made to behave will be made to behave the way Italy just did it, one enforceable rule at a time, about data and disclosure and accountability, by regulators who were unfashionable a year ago and are about to be the most important people in the room. Stop waiting for the technology to become safe. Start writing down what you will and will not let it touch. That list is the only pause you actually control.

The Ledger

The pause letter (unenforceable) and Italy's ban (enforceable) frame the fortnight. Regulation, not a moratorium, is how this gets slowed, and it will be about data and disclosure first. Watch the EU AI Act, now moving through Brussels.

Data centers & power

Quiet on the front page. Under it, every one of these models is training on Nvidia hardware in a data centre somewhere, and the demand curve is bending upward. Nobody is pricing the electricity yet.

Rates

The banking scare has done some of the Fed's tightening for it; markets read the March turmoil as worth roughly a rate rise. The March jobs and inflation prints land after this issue.

Real estate

US 30-year mortgage 6.28 percent (Freddie Mac, April 6). Commercial property, especially offices, is the slow-motion worry behind the regional-bank story.

India tech

India is days from being declared the world's most populous country. The services industry's pitch, a young workforce at scale, has never had a better number behind it, and never a harder question about whether the work it does is the work AI is about to automate.

What we called wrong

In No. 4 we wrote that Google had shown "what it looks like when the confident machine is wrong in public with your name on it." Fair, but we implied it might cost Google the race. A fortnight of Bard improving and Bing plateauing says we should not have written the obituary in February.

The Back Page

The Spaniard in the green jacket

Jon Rahm won the Masters on the evening of April 9, closing with a four-under 68 at Augusta National to finish twelve under and four clear, on the eleventh anniversary, almost to the day, of Seve Ballesteros's death. Rahm, from the Basque Country, kept a photograph of Seve in his mind all week and said so afterwards, in tears. He had started the tournament by four-putting the first green on Thursday and made a birdie and an eagle in the next three holes, which is the whole of golf in four holes: the game punishes you and then, before you have finished being punished, offers you everything back. He is the second Spaniard to wear the green jacket after Ballesteros and José María Olazábal, and he putted the last one in as the Georgia light went gold.