The venerable magazine Scientific American is facing an existential crisis, having been sold by Springer Nature after annoying the Trump administration. CC-licensed photo by Don Harder 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. Unprintable. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Scoop: unvetted AI agents are accessing US bank accounts • Axios
Madison Mills:
»
Major U.S. banks are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents attempting to take actions that appear to be independent of user instructions, sources tell Axios.
AI agents are becoming a nightmare for America’s banks, making their way into banking systems faster than financial institutions can respond.
The difficulty in distinguishing between a human and an unauthorized bot is forcing banks to block the technology and rapidly rewrite their terms of service.
Banks including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and others are racing to contain agent risks. Some are aware of unauthorized access, people familiar with the issue said.
These are generally situations when an agent or AI assistant accesses a human’s account with the user’s permission and login credentials. Even if permitted by the user, this goes against the guidelines of most banks requiring that humans log in to their own accounts.
This presents significant security risks: Banking executives fear the steps agents with login access might take, especially if a user has granted purchase authority or the ability to shift funds.
In other instances, some banks have tracked activity when agents — programmed to fulfill requests or complete tasks on a user’s behalf — contact call centers or appear to take actions independent of user instructions, the people said.
Bank executives described the situation as horrifying, with one official saying the industry has grave concerns about agents gaining access to customer accounts and taking steps that could impact user finances.
«
Morgan Stanley, JP Morgan, BoA and Citibank are all blocking or restricting agents as of now. But you have to think that some people have more money than sense, or that the latter has been wiped away by the excitement of running an agent. Has nobody watched Silicon Valley?
unique link to this extract
April 2026: Vulnerability research is cooked • A Final Ward
Thomas Ptacek:
»
You can’t design a better problem for an LLM agent than exploitation research.
Before you feed it a single token of context, a frontier LLM already encodes supernatural amounts of correlation across vast bodies of source code. Is the Linux KVM hypervisor connected to the hrtimer subsystem, workqueue, or perf_event? The model knows.
Also baked into those model weights: the complete library of documented “bug classes” on which all exploit development builds: stale pointers, integer mishandling, type confusion, allocator grooming, and all the known ways of promoting a wild write to a controlled 64-bit read/write in Firefox.
Vulnerabilities are found by pattern-matching bug classes and constraint-solving for reachability and exploitability. Precisely the implicit search problems that LLMs are most gifted at solving. Exploit outcomes are straightforwardly testable success/failure trials. An agent never gets bored and will search forever if you tell it to.
Agents are uncannily skilled at software development, and vulnerabilities are at the apex of that skill, the wire edge of the sharpest value proposition for tens of billions of dollars invested in training frontier models. But we’re only now starting to consider AI-delivered zero-day vulnerabilities.
…In 2025, the vendor metagame for high-end exploit development was to buy a crate of Vyvanse and Provigil for a bunch of European Zoomers and have them stay awake for 4 days straight studying the memory lifecycle of CSS stylesheet objects. Vendors won’t need chemical accelerants (and Zoomers) much longer. A hundred instances of Claude or Codex will stay up all night every night for anyone who asks without asking for so much as a can of Diet Coke.
«
And so it is coming to pass, six months later.
unique link to this extract
The fall of “Scientific American” • The Nation
Emmet Fraizer:
»
In March of 2026, in a nondescript, 46th-floor office overlooking lower Manhattan, about two dozen members of Scientific American’s editorial staff had gathered in person and online to talk openly about their approach to political coverage.
David Ewalt, the editor in chief, wanted them all to understand something: The magazine was not “kowtowing” to the Trump administration. In fact, according to a recording of the meeting obtained by The Nation, he asked staff present to “raise your hand if you’ve ever heard me say not to cover a political story.”
Two people raised their hands.
For well over a year, rumours had been circulating among the staff about top-level editors’ rejecting pitches they saw as too political, controversial articles triggering angry phone calls from Scientific American’s corporate owner, and even, on at least one occasion, ownership forcing editors to blunt language critical of the Trump administration.
Then there was what happened to the opinion department. In the first months of the new administration, it had been publishing about 15 pieces a month, many of which criticized Trump’s attacks on scientific research. But after the department published a particularly provocative column in May 2025—“Science Tells Us the U.S. Is Heading toward a Dictatorship”—management, in what one staffer would later call an “unprecedented” move, had told the department to “pause” publishing new articles until the new editor in chief started.
«
Fraizer gets into the nitty-gritty inside the magazine, and its sale by Springer Nature (apparently exasperated by articles like that May 2025 one; which, honestly, would be fine as a general politics article, but to pretend it’s science (“but it’s pOLiTiCaL science!”) is daft) to a little-known publisher.
But what is omitted is the wider picture: that the market for a science magazine had been dying for years, so what was Scientific American’s USP? In the UK, New Scientist is hard science, feature and news article length. Why would you buy or subscribe to Scientific American, though?
unique link to this extract
Fixing photosynthesis: the innovation that could feed the world • Institute of Economic Affairs
Matt Ridley:
»
There are several ways in which photosynthesis wastes energy. It cannot use infrared, it reflects away green light (hence the colour of plants) and it wastes much of the energy in blue photons. But the big inefficiency comes with the design of a critical protein machine called rubisco. This 18,000-atom molecular device at the heart of photosynthesis struggles to distinguish oxygen from carbon dioxide. So almost half the time it undoes its own work, unfixing carbon and wasting energy.
The reason it does this is probably a historical accident. When rubisco first emerged three billion years ago there was no oxygen in the atmosphere, so “photorespiration” was not a problem. And the reason life has failed to solve it since is that all ways of making the enzyme better at discriminating between the two gases tend to slow down the reaction. Most plants, known as C3 plants, simply tolerate photorespiration and fine-tune rubisco to be as discriminating as possible without slowing things down too much.
About 35 million years ago, some plants came up with a neat workaround, called the carbon-concentrating mechanism, whereby they gather carbon dioxide up in special compartments and feed it to rubisco. This is a bit like James Watt’s separate condenser for the steam engine: it wastes some energy, but not as much. These plants are known as C4 plants, after the four-carbon sugar they make as an intermediate step, and they thrive in warmer, drier conditions.
Various versions of the C4 mechanism were invented around 65 times independently. Maize and sugar cane are good examples: they grow faster than wheat and rice but only in warm weather. Around 3% of plant species are C4 plants, but they so dominate tropical ecosystems that they account for nearly a quarter of all plant photosynthesis. An Oxford-led international project is exploring how to turn rice into a C4 plant, with the goal of increasing rice yields by 50%, but progress has been slow.
Why not redesign rubisco itself? Short answer: because if it were easy, Mother Nature would have found a way. She does not like leaving opportunities like that lying around any more than we do. But a longer answer is that she sometimes gets stuck in the “path dependence” of evolution, a captive of history. You can’t get there from here by evolution – but maybe you can by innovation.
«
Obviously Ridley thinks we should use AI to do this; it’s only 1% effective at present. It would have been helpful to know what 10% effective would look like: would we start clearing the excess CO2 from the atmosphere?
unique link to this extract
The art of the deal: Jared Kushner’s hacked phone revealed his secret negotiations— including more than $700m in transfers • Vanity Fair
Matthew Cole:
»
In late spring 2020, a source I knew to be reliable called me with a story. The Dutch government had hacked into Jared Kushner’s phone and monitored him as he accessed a Dubai-based bank account. What was odd, the source said, was that the money in the account—hundreds of millions of dollars—had come from a bank in Azerbaijan. Most importantly, Kushner appeared to have transferred significant amounts of money from the Dubai account to other individuals, including a warlord, a mercenary and a one-time hit man. The story sounded preposterous, I told my source. And yet.
Six months later, I found myself speaking with someone else about the UAE government and their use of mercenaries in conflicts around the Middle East and Africa when I asked, on a whim, whether they knew anything about Kushner and bank transfers from a UAE account. My acquaintance grew quiet. Then they replied, “Yes.”
Over the next four years, I came to learn that, sometime in late 2017, the Netherlands Defense Intelligence and Security Service (MIVD) made a startling discovery. A General Staff of the Armed Forces of the Russian Federation (GRU) operative, a dark web hacker specializing in stolen bank data, was traveling to the UK for an espionage mission. They reached out to the British government for permission to apprehend him when he arrived. The British agreed.
…Later that year, the Russian operative gave the Dutch spies access to one of the GRU’s most sensitive assets. Located in Rostov, Russia, the GRU’s hacking and cyber team operated a remote server on which they stored some of their bounty. During their search through these secret files, the Dutch found something astonishing: screenshots of Jared Kushner’s iPhone.
In The Hague, the MIVD hacking team tried to reverse engineer the Russian penetration of Kushner’s iPhone. Kushner owned what was then the latest model, the iPhone X. …By early 2018, the MIVD had found a vulnerability in the iPhone that gave them access. The security exploit wasn’t perfect, however. Some malware allows a hacker to turn a victim’s phone into a microphone and video camera, turn the phone on or off, and pull and add files. The Dutch solution did not provide that level of access. Instead, when Kushner used the phone, they could see his screen. Almost as soon as they got inside, the intelligence service began capturing every image of Kushner’s screen as he used it, filing their own cache of surveillance on their internal computer servers.
«
Should be fun after the midterms when a Democratic Congress asks Cousin Greg to come and testify. (Kushner denies it all.)
unique link to this extract
The iPhone Duo • Daring Fireball
John Gruber reviews (at length) Apple’s foldable. These are the first four paragraphs:
»
The very nature of the iPhone Duo raises numerous obvious questions about what it’s like to use. Does it not feel thick and heavy, awkward both to pocket and unpocket? Doesn’t the outer (cover) screen feel small and cramped, with an unfamiliar squat-rather-than-tall layout for UI controls in every app you use? Isn’t it weird that when you open it up to use the inner screen, that all the hardware buttons on the edges of the device are in the “wrong” places for a tablet? For anyone who’s gotten used to Face ID over the last decade, is it not frustrating to go back to Touch ID? And what about third-party apps that haven’t been updated to support the Duo’s novel displays — do they not suck to use?
Clearly, the point of a Duo review is to answer these questions, which really boil down to one question: how did Apple solve these problems?
The answer is: they didn’t.
All of these problems are real, and each of them detracts from the experience of using the Duo. Some of them have been mitigated through thoughtful design and clever engineering. Some of them might be solved in future iterations of the device. Some of them, though, I believe are endemic to the very idea of a foldable, and cannot be solved.
«
It’s hard not to feel that Apple has done this because it feels it ought to – that letting Samsung and Huawei and a few others do foldables, but not taking some of that market itself, was leaving money on the table – but the expense of doing this design, the low yields (due to the challenge of making the hinge) and the inherent problems of foldables, as Gruber explains in detail, suggest it’s a form that will not expand beyond a few% of the smartphone market.
unique link to this extract
Indian firm eMudhra exits global digital certificates business after vulnerability disclosure • Reuters via MSN
Abhirami G:
»
Indian digital trust services provider eMudhra on Thursday said it would voluntarily withdraw from the global SSL/TLS certificate business, following an incident in which its digital certificates were found to have vulnerabilities.
The disclosure of the vulnerabilities led to internet browser majors such as Google, Mozilla, Apple and Microsoft to block digital certificates issued by eMudhra, and sparked larger concerns among cybersecurity professionals around internet security and privacy.
SSL and TLS are protocols used to secure data sent between web browsers and servers, ensuring that private information cannot be intercepted by bad actors. eMudhra issues these certificates to digital businesses and government agencies, including for some online public services such as India’s income tax portal.
An independent cybersecurity researcher found over the weekend that several key certificates issued by eMudhra, used to secure websites, software, and online communications, were weak and vulnerable to attack.
On Wednesday, India’s Computer Emergency Response Team also issued an advisory saying that some of eMudhra’s intermediate certificates were compromised.
The company, in its statement on Wednesday, said it withdrew eight certificates it had issued, and that its infrastructure was not compromised and no data was breached.
The level of social media amplification around the incident has brought scrutiny that is out of proportion with the significance of the vertical in the company, eMudhra said in its follow-up statement, adding that the global SSL/TLS certificate business contributes 1% of its revenue.
«
Well, that was quick. The suspicion is that Google had intimated that the Chrome browser would *never again* trust certificates issued by eMudhra. And everyone knows that if you don’t work on Chrome, there’s no point being online.
unique link to this extract
Mercury to gold: Marathon Fusion claims 10x CERN record, but can ‘modern alchemy’ work? • BusinessToday
»
A San Francisco-based fusion startup says it has achieved something that sounds straight out of medieval alchemy: turning mercury into stable gold using fusion-generated neutrons. But beyond the headline-grabbing claim lies a much bigger ambition — using the same technology to produce scarce medical isotopes for cancer diagnosis and treatment, while developing the engineering capabilities needed for future fusion power plants.
Marathon Fusion says its experiments have demonstrated mercury-to-gold transmutation at a rate it claims is 10 times the record set by CERN in 2025. The company has also reported experiments involving isotope separation and the production of radioactive copper isotopes from zinc.
The claims have attracted attention from technology and investment enthusiasts, including questions about whether artificially produced gold could eventually challenge the economics of the precious metal. However, the distinction between demonstrating a nuclear reaction and producing gold commercially at scale is crucial.
The process is called nuclear transmutation. Unlike chemical reactions, which rearrange atoms without changing their nuclei, nuclear transmutation changes an atom’s nucleus to produce a different element or isotope.
Marathon Fusion’s proposed method uses neutrons generated by a deuterium-tritium fusion reaction to transform mercury-198 into mercury-197. The latter is radioactive and can decay through electron capture into gold-197, the stable isotope of gold.
• Start with mercury: The proposed process uses mercury-198, a naturally occurring stable isotope of mercury, as its starting material
• Use fusion-generated neutrons: High-energy neutrons strike mercury-198 nuclei, triggering a reaction that ejects two neutrons and produces mercury-197
• Produce gold: Mercury-197 decays into stable gold-197 [with a half-life of 65 hours]. The challenge is to achieve this at an industrial scale while safely managing radioactive byproducts.«
That’s the challenge, yes. Oh – you wondered how much gold CERN generated? 10 picograms. That’s 10 x 10^-12 grams. So they could make 100 x 10^-12 grams. The world gold market might be safe for a bit.
unique link to this extract
Exclusive: Trump called Musk and asked him to lift Ukraine Starlink restrictions • WSJ
Alexander Ward and Lara Seligman:
»
President Trump called Elon Musk on Wednesday and asked him to lift restrictions on a military-grade version of Starlink in Ukraine that the SpaceX chief had throttled, US officials said.
The system was working more normally Thursday, shortly after their conversation, US and Ukrainian officials said. Some Ukrainian officials said the system wasn’t restored to full functionality but was working at higher speeds than under the cap.
Trump told Musk he was frustrated the system’s speeds remained capped at around 100 miles an hour, officials said. The speed limit rendered some of Ukraine’s offensive drones and interceptors powerless against Russian attacks. Their conversation followed a Wall Street Journal report revealing that Musk ordered the slowdown, though SpaceX and the White House have differing accounts about whether the world’s richest man acted alone or at the administration’s behest.
Trump has told aides and allies this week that he wanted to reverse the slowdown of Starlink, which Ukraine’s forces use to defend themselves along the front line, the officials said.
“Removal of the throttle limit will do much to enable Ukraine’s strike capabilities at operational depths and its current effort to scale up deployment of jet-powered interceptors,” said Michael Kofman, an expert on the war at the Carnegie Endowment for International Peace. Kofman added that he had heard from Ukrainian military units that early tests suggest the system is now working more regularly.
«
Possibility 1: Trump told Musk to slow it down. Possibility 2: Musk decided to slow it down. Possibility 3: it was an accident putting Ukraine into the “same speed as any doofus” category.
Event 1: stories appear saying Musk slowed it down (not sourced to Musk), and then that Trump told Musk to speed it back up (again, not sourced to Musk).
Given Trump’s posture towards Ukraine, I don’t find Possibility 2 very likely, though 3 is, and 1 is.
The measurement in miles per hour is because the satellites have to hand over the signal between each other for a moving connection; 100mph is the standard for the rest of us (since March 2026), while the system can cope with 450mph.
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
