Chess computers have radically changed the complexion of the game. Could AI authors do the same for science? And is that good or bad? CC-licensed photo by Ryan Somma 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 10 links for you. Check, mate. 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 possible US government iPhone-hacking toolkit is now in the hands of foreign spies and criminals • WIRED
Andy Greenberg:
»
An iPhone-hacking technique used in the wild to indiscriminately hijack the devices of any iOS user who merely visits a website represents a rare and shocking event in the cybersecurity world. Now one powerful hacking toolkit at the center of multiple mass iPhone exploitation campaigns has taken an even rarer and more disturbing path: It appears to have traveled from the hands of Russian spies who used it to target Ukrainians to a cybercriminal operation designed to steal cryptocurrency from Chinese-speaking victims—and some clues suggest it may have been originally created by a US contractor and sold to the American government.
Security researchers at Google on Tuesday released a report describing what they’re calling “Coruna,” a highly sophisticated iPhone hacking toolkit that includes five complete hacking techniques capable of bypassing all the defenses of an iPhone to silently install malware on a device when it visits a website containing the exploitation code. In total, Coruna takes advantage of 23 distinct vulnerabilities in iOS, a rare collection of hacking components that suggests it was created by a well-resourced, likely state-sponsored group of hackers.
In fact, Google traces components of Coruna to hacking techniques it spotted in use in February of last year and attributed to what it describes only as a “customer of a surveillance company.” Then, five months later, Google says a more complete version of Coruna reappeared in what appears to have been an espionage campaign carried out by a suspected Russian spy group, which hid the hacking code in a common visitor-counting component of Ukrainian websites. Finally, Google spotted Coruna in use yet again in what seems to have been a purely profit-focused hacking campaign, infecting Chinese-language crypto and gambling sites to deliver malware that steals victims’ cryptocurrency.
Conspicuously absent from Google’s report is any mention of who the original surveillance company “customer” that deployed Coruna may have been. But the mobile security company iVerify, which also analyzed a version of Coruna it obtained from one of the infected Chinese sites, suggests the code may well have started life as a hacking kit built for or purchased by the US government.
«
As Benedict Evans observed, this illustrates Apple’s point that if you build a backdoor only for the government, it will leak from the “good guys” you hope are in government to the bad guys who definitely exist outside it.
unique link to this extract
Ars Technica fires reporter after AI controversy involving fabricated quotes • Futurism
Maggie Harrison Dupré:
»
The Condé Nast-owned Ars Technica has terminated senior AI reporter Benj Edwards following a controversy over his role in the publication and retraction of an article that included AI-fabricated quotes, Futurism has confirmed.
Earlier this month, Ars retracted the story after it was found to include fake quotes attributed to a real person. The article — a write-up of a viral incident in which an AI agent seemingly published a hit piece about a human engineer named Scott Shambaugh — was initially published on February 13. After Shambaugh pointed out that he’d never said the quotes attributed to him, Ars‘ editor-in-chief Ken Fisher apologized in an editor’s note, in which he confirmed that the piece included “fabricated quotations generated by an AI tool and attributed to a source who did not say them” and characterized the error as a “serious failure of our standards.” He added that, upon further review, the error appeared to be an “isolated incident.” (404 Media first reported on the retraction.)
Shortly after Fisher’s editor’s note was published, Edwards, one of the report’s two bylined authors, took to Bluesky to take “full responsibility” for the inclusion of the fabricated quotes.
In the post, Edwards said that he was sick at the time, and “while working from bed with a fever and very little sleep,” he “unintentionally made a serious journalistic error” as he attempted to use an “experimental Claude Code-based AI tool” to help him “extract relevant verbatim source material.” He said the tool wasn’t being used to generate the article, but was instead designed to “help list structured references” to put in an outline. When the tool failed to work, said Edwards, he decided to try and use ChatGPT to help him understand why.
“I should have taken a sick day because in the course of that interaction, I inadvertently ended up with a paraphrased version of Shambaugh’s words rather than his actual words,” Edwards continued.
«
Wow. First of all: Edwards was working and being expected to work while he was sick? And: the single mistake leads to him being fired? Unless he was on some number of warnings already, this seems disproportionate. Could Condé Nast perhaps think about not making people work when they’re ill? Meanwhile: Edwards has always seemed to me an effective reporter. Let’s hope he gets picked up soon. (On Bluesky, he said that he has been struggling for a long time with Covid infections and their effect.) This stands in stark contrast to the next story…
unique link to this extract
Staff journalists sacked and misleadingly replaced with AI writers • Press Gazette
Rob Waugh:
»
A network of prominent gaming sites has fired multiple human staff in recent days and misleadingly replaced them with AI writers, complete with fake pics and biogs.
UK-based The Escapist, Videogamer and Esports Insider were taken over by SEO agency Clickout Media in recent months, with up to 20 staff believed to have been fired.
Videogamer staff and freelances, who did not want to be named and said the company still owed them money, said late last year the new owner began to load the sites with AI-written stories about casinos.
Then this year, budgets were frozen and staff were told to reapply for new roles where they would be training AI “writers”.
Videogamer has been in business more than 20 years. At the top of every story is the following statement: “You can trust VideoGamer. Our team of gaming experts spend hours testing and reviewing the latest games, to ensure you’re reading the most comprehensive guide possible. Rest assured, all imagery and advice is unique and original.”
New writer on the site Brian Merrygold has a byline picture which is AI generated (according to checking service IdentifAI). His biog is also entirely AI generated (per text checking service Pangram), as are his articles.
His biog states that he is “an experienced iGaming and sports betting analyst” and a “lifelong gamer at heart”.
…Clickout Media describes itself as a “PR and marketing agency” but has a history of acquiring gaming and tech sites including Techopedia and Adventure Gamers, firing staff and replacing them with seemingly automated content around casinos and cryptocurrency.
«
For some reason I’m thinking of parasitic wasps which lay their eggs in other insects, which then die as they’re eaten from inside. (That idea was the inspiration for Alien.)
unique link to this extract
How the experts figure out what’s real in the age of deepfakes • The Verge
Jess Weatherbed:
»
Step one: look very, very closely. When unverified images of Venezuelan leader Nicolás Maduro suddenly proliferated on social media after his abduction by the US in January, The [New York] Times’s Visual Investigations team jumped into action. They scrutinized the images for visual inconsistencies “that would suggest they were not authentic” — such as one example that featured an aircraft with odd-looking windows.
This wasn’t enough to definitively prove the pictures were fake. “But even the remote chance that the images were not genuine — coupled with the fact they came from unknown sources, and details like Mr. Maduro’s clothing being different between the two images — was strong enough to disqualify them from publication,” The Times’s photography director Meaghan Looram said in the article.
We’re mostly past the days of identifying AI-generated deepfakes by counting how many fingers a person has, but there are usually still subtle indicators — for instance, check the architecture and figures in the backgrounds for unexplained oddities.
Step two: consider the source and its reputation. One image of Maduro that The Times did publish — showing the Venezuelan leader in custody — came from President Donald Trump’s Truth Social account. That doesn’t mean Trump or any other government official is a reliable source — he has a habit of disseminating AI fakery online, and the integrity of government handouts generally can be difficult to establish. Authenticity concerns were also flagged for the image in question, regarding its poor quality and unusually cropped dimensions.
“In this case, the president’s Truth Social post itself was newsworthy, even if we had no surefire way to confirm that the image was authentic,” said Looram. But it was published on The Times’ homepage as part of a screenshot of Trump’s full post, not in isolation. “Displaying it in context means that, if the image proves to be inauthentic in some way, we will not have presented it as a legitimate news photo, but rather as a communication from the President.”
«
In other words: it transfers the burden of truth onto those being quoted, rather than the publisher. Lots of people seem to find this maddening and insist that the publication should somehow magically determine for itself what is true for everything and never get it wrong. Which only suggests that they’ve never had to consider how much work is involved in producing a newspaper or its website, and how much reliance one has to put on trusting sources.
unique link to this extract
The legibility problem • Asimov
Matthew Carter is a computational biologist:
»
The last World Computer Chess Championship was held in October 2024. It ended because, in the words of its organizers, “top programs are unbeatable by humans; making them stronger has no real research value.” Its mission, after half a century of effort, was complete.
Chess engines now sit at the center of how the game is played. Engine evaluations hover over boards during livestreams and post-game analyses, and players study engine lines extensively to improve their own play. As a result, top grandmasters are stronger than any previous generation. But these gains come with a concession: Top engines often play in a way “too far past the human frontier” to fully understand. In other words, we trust chess engines because they are unquestionably better than we are, even when their decisions make no sense to us.
A similar imbalance may also emerge in scientific discovery. Researchers are building systems able to navigate the full arc of scientific inquiry with increasing autonomy, including proposing hypotheses, designing experiments, and evaluating results. The question, however, is whether we will cede the same authority to AI scientists as we have done in chess. And, if so, what will happen to science when AI models produce results beyond our ability to understand?
I call this the “legibility problem,” the risk that AI-generated scientific knowledge becomes incompatible with human understanding, and think it will define the next era of science. The knowledge AI systems generate may be expressed in concepts that do not map onto our own, communicated in ways optimized for other AIs rather than for human investigators.
…Take, for instance, the diabetes drug metformin, ingested by millions of people for over 70 years. Despite its success, we still do not fully understand its mechanism of action. But metformin did not appear out of nowhere. It emerged through a long chain of human experimentation, from herbal medicine and chemical purification to animal studies and clinical trials.
AI-generated science may not follow this pattern. AI scientists will have no intrinsic reason to work within our existing conceptual categories, just as superhuman chess engines have no intrinsic reason to explain their choice of moves. In fact, if we truly want AI scientists to make breakthroughs, some loss of legibility may be inevitable. The chief risk is that discoveries become effectively stranded, buried in a volume of AI-generated output no human institution is equipped to parse or implement.
«
This is on a par with humans’ struggle to understand Go AI programs’ moves – but much, much harder.
unique link to this extract
Apple accidentally leaks “MacBook Neo” • MacRumors
Joe Rossignol:
»
Apple appears to have prematurely revealed the name of its rumoured lower-cost MacBook model, which is expected to be announced this Wednesday.
A regulatory document for a “MacBook Neo” (Model A3404) has appeared on Apple’s website. Unfortunately, there are no further details or images available yet.
While the PDF file does not contain the “MacBook Neo” name, it briefly appeared in a link on Apple’s regulatory website for EU compliance purposes.
The lower-cost MacBook is rumoured to feature an iPhone chip like the A18 Pro or A19 Pro, rather than an M-series chip, as well as a 12.9in display. It has also been rumored that this MacBook will come in fun color options, like yellow, green, blue, and/or pink, and the “MacBook Neo” name certainly sounds fun.
“MacBook Neo” would slot in below the MacBook Air in the Mac lineup, but its starting price remains to be seen, with estimates ranging from $599 to $799.
«
There have been mistakes like this in the past, where Apple has inadvertently revealed details about upcoming products on its website. People have been fired for those mistakes, certainly during the Steve Jobs era. Of course, you’d need the very finest search technique in the world to find them.
Or you could ask ChatGPT, which says that the “lampstand” iMac G4, iPhone 3GS, 2010 Mac Pro and Mac mini updates, and Airport Extreme were all shown on its site by mistake. I honestly can’t recall, but the Mac Pro was surely one of them. (Finding a reference to a 16-year-old event, of course, is impossible even so.)
I don’t think Apple did this on purpose – though if it did, this would be a brilliant way to leak the name but not the form factor.
unique link to this extract
Why Katie Miller and other MAGA influencers suddenly love solar power • The Washington Post
Evan Halper:
»
A growing number of prominent Trump allies — including former House speaker Newt Gingrich, veteran strategist Kellyanne Conway and GOP pollster Tony Fabrizio — are promoting solar as electricity demand surges and energy affordability climbs the list of voter concerns.
Their clean energy advocacy may be having an impact, as the White House signals it is reconsidering power from the sun. The tone of Trump himself has even changed.
In an interview, [MAGA influencer Katie] Miller said solar is crucial to delivering on the right’s energy and AI dominance agenda. “Look at what Australia did,” she said. “Solar solved their rolling blackout issues. President Trump has prioritized lowering the cost of energy for the American people … I am simply advocating that solar can and should be a driver of the solution.”
Asked if she is getting paid for her advocacy, like some other MAGA heavyweights promoting solar, Miller would not comment. Regardless, these full-throated endorsements of a renewable energy source that has been much maligned by Trump and his advisers represents a departure from what had been a pillar of the MAGA energy agenda.
It reflects a realization taking hold more broadly among Republicans that solar power — long embraced by liberals — is increasingly indispensable to America’s bid to dominate AI, close a yawning “electron gap” with China and contain runaway residential electricity costs. These conservatives describe it as crucial to U.S. competitiveness, the grid’s reliability and their own movement’s political survival. Climate change rarely enters the conversation.
«
No need to mention climate change; the cost and speed of installation arguments suffice. Plus anxiety on MAGA’s part about losing out to China, which installed more solar in 2025 than the US has in its entire history.
Also, of course, solar panels tend not to travel through the Straits of Hormuz.
unique link to this extract
ChatGPT uninstalls surged by 295% after DoD deal • TechCrunch
Sarah Perez:
»
U.S. app uninstalls of ChatGPT’s mobile app jumped 295% day-over-day on Saturday, February 28, as consumers responded to the news of OpenAI’s deal with the Department of Defense (DoD), which has been rebranded under the Trump administration as the Department of War.
This data, which comes from market intelligence provider Sensor Tower, represents a sizable increase compared with ChatGPT’s typical day-over-day uninstall rate of 9%, as measured over the past 30 days.
Meanwhile, U.S. downloads for OpenAI competitor Anthropic’s Claude jumped up by 37% day-over-day on Friday, February 27, and 51% as of Saturday, February 28, after the company announced that it would not partner with the U.S. defense department. Anthropic said it was not able to agree on the deal terms over concerns that AI would be used to surveil Americans and be used in fully autonomous weaponry, which AI is not yet ready to do safely.
A set of consumers seemingly favored Anthropic’s position on the matter, the data suggests.
In addition, ChatGPT’s download growth was impacted by the news of its DoD partnership, with its U.S. downloads dropping by 13% day-over-day on Saturday, shortly after the news of its deal went public. Those downloads continued to fall on Sunday, when they were down by 5% day-over-day. (Before the partnership was announced, the app’s downloads had grown 14% day-over-day on Friday.)
These rapid changes were also reflected in Claude’s App Store ranking, as the app hit No. 1 on the U.S. App Store on Saturday, where it continues to sit as of Monday, March 2. That’s a jump of over 20 ranks compared with roughly a week before (February 22, 2026).
«
Sensor Tower’s methodology might be a little off, though the trend is clear enough (and the App Store rankings probably reflect the picture more accurately). But all the percentages, rather than hard numbers, are annoying.
unique link to this extract
Inside the plan to kill Ali Khamenei • Financial Times
Mehul Srivastava, James Shotter, Neri Zilber and Steff Chávez:
»
When the highly trained, loyal bodyguards and drivers of senior Iranian officials came to work near Pasteur Street in Tehran — where Ayatollah Ali Khamenei was killed in an Israeli air strike on Saturday — the Israelis were watching.
Nearly all the traffic cameras in Tehran had been hacked for years, their images encrypted and transmitted to servers in Tel Aviv and southern Israel, according to two people familiar with the matter.
One camera had an angle that proved particularly useful, said one of the people, allowing them to determine where the men liked to park their personal cars and providing a window into the workings of a mundane part of the closely guarded compound.
Complex algorithms added details to dossiers on members of these security guards that included their addresses, hours of duty, routes they took to work and, most importantly, who they were usually assigned to protect and transport — building what intelligence officers call a “pattern of life”.
The capabilities were part of a years-long intelligence campaign that helped pave the way for the ayatollah’s assassination. This source of real-time data — one of hundreds of different streams of intelligence — was not the only way Israel and the CIA were able to determine exactly what time 86-year-old Khamenei would be in his offices this fateful Saturday morning and who would be joining him.
Nor was the fact that Israel was also able to disrupt single components of roughly a dozen or so mobile phone towers near Pasteur Street, making the phones seem as if they were busy when called and stopping Khamenei’s protection detail from receiving possible warnings.
Long before the bombs fell, “we knew Tehran like we know Jerusalem”, said one current Israeli intelligence official. “And when you know [a place] as well as you know the street you grew up on, you notice a single thing that’s out of place.”
«
This is only the prelude, but even this shows the huge amount of subterfuge and secrecy – maintained for years – that was necessary to have a potentially winning move in a war.
unique link to this extract
Podcasts lead AM/FM in spoken-word listening, marking a first • Edison Research at SSRS
Edison Research:
»
In 2015, AM/FM radio accounted for 75% of the time Americans spent with spoken-word audio sources. AM/FM radio was not only the most dominant spoken-word audio listening platform, but it was fully sixty-five percentage points higher than podcasts, which accounted for 10% of listening time back then.
Quarter by quarter and year over year, time spent using AM/FM radio to listen to spoken-word audio has declined significantly and shifted to time spent with podcasts. As of Q4 2025, 40% of time spent listening to spoken-word is now spent with podcasts and 39% of time is spent with AM/FM radio. Not only does radio not beat podcasts by a significant margin, it now trails the on-demand platform for spoken-word audio listening.
«
I suspect this is indicative of how people want to customise their experience; though additionally the figures for AM/FM plus podcasts has remained at around 85%-80% (with a dip during the Covid period). That suggests these are commuting experiences, with digital radio and audiobooks making up the remaining time.
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
The Overspill: when there's more that I want to say 