As you get older, do you start to slow down.. on the road? If so, why? CC-licensed photo by Brian Suda 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. Stay in your lane. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Amazon security research reportedly led to the White House’s Anthropic Fable ban • The Verge
Terrence O’Brien:
»
According to the Wall Street Journal, the export control directive that led to Anthropic cutting off access to Fable 5 and Mythos 5 was triggered in part by cybersecurity research from Amazon and conversations between CEO Andy Jassy and the White House. According to the report, the paper from Amazon claims that, through a series of prompts, it was able to get Fable 5 to serve up information that could be used in cyberattacks. Amazon has yet to respond to a request for comment.
Shortly after Jassy shared the company’s findings with the government, it made the call to block its use by foreign nationals. Complicating this issue is that many of Anthropic’s researchers are foreign-born, meaning they were barred from accessing their own product.
In a statement, Anthropic disputed the government’s characterization of the issue as a “jailbreak.” It argued that many of the same vulnerabilities could be discovered using other publicly available models, including GPT 5.5. Some security researchers appear to back the company’s interpretation. Katie Moussouris, the founder and CEO of LutaSecurity posted on BlueSky that “I’ve seen the paper. It’s not a jailbreak.” Former Commerce Department official Kate Koren speculated to the WSJ that the White House’s dislike of Anthropic may have influenced the decision.
Anthropic and the Trump administration have been at odds for some time over the company’s refusal to allow its AI to be used for mass surveillance of Americans or to power lethal autonomous weapons. In February, Trump instructed federal agencies to stop using Anthropic’s AI. And just hours later, Secretary of Defense Pete Hegseth designated the company a supply chain risk.
«
This beef between Amazon (has US defence contracts) and Anthropic (presently shut out of US defence contracts) has the potential to become really explosive. Would Anthropic deny Amazon access to its models? It probably has to to any Amazon employees who are not US-born. But how do you do that? Better deny it to everyone. The problem with that, though, is that you can’t be absolutely sure that someone is a US citizen even at a US company. Ironically, some of Anthropic’s top researchers are now disallowed from accessing their own product because they’re not US citizens.
unique link to this extract
In age of AI, world’s leading deepfake expert no longer trusts his own eyes • The New York Times
Eli Saslow:
»
[Hamid Farid’s wife Emily] Cooper was a leading vision scientist at Berkeley. She researched how humans perceive reality, while her husband investigated how that reality could be faked. They’d collaborated on studies about deepfakes, but in the last months that research had begun to follow them home. Instead of dealing with one case every few weeks, Farid was working as an adviser and an expert witness, juggling up to a dozen cases each day. For the first time in his career, he’d become not just an analyst but also a victim, when someone spoofed his cellphone number and used AI to clone his voice. The hacker made calls to one of Farid’s colleagues on a sensitive case, impersonating Farid and pressing for confidential information. Now Farid and Cooper had decided never to take their identity for granted. They invented a safe word to confirm they were real at the start of any sensitive phone call.
Farid glanced at his phone and saw a new email: “I’m fact-checking this viral video of a mother and child approaching a flag-draped coffin that we suspect is AI generated,” it read. He set the phone back down and looked out from the porch at the Berkeley Hills, the San Francisco Bay, and the sun setting over the Golden Gate Bridge.
“I can’t stand this place anymore,” he said. “These major tech giants will burn everything to the ground as long as they’re making a profit. They’re not interested in anything that’s going to slow them down.”
“It makes me anxious for our students,” Cooper said. “It’s starting to scare me.”
«
(Gift link.)
unique link to this extract
I was scammed buying GLP-1s online. I’m not alone • WIRED
John Semley:
»
In March, I signed up for a service called FitRx—which also does business under the name Zealthy, Inc.—mostly to see the cost of compounded Zepbound. The website advertises an offer of $135 for a month’s supply. I registered, paid the membership fee, and answered some basic questions about my weight, medical history, and activity level. When I woke up seven hours later, I found a detailed message from a physicians’ assistant—who I did not communicate with directly—laying out my new “plan.” I also found I was charged $866 and had been sent a three-month supply of tirzepatide vials that I had been prescribed, despite neither asking for or needing them. No effort whatsoever was made to check if these unwanted medicines were covered by insurance.
I spent the better part of the next week arguing with customer service reps at FitRx and Zealthy about their cancellation rules. (With FitRx, this almost always involved dialing in to be cheerily greeted by the same dude, who would only identify himself as “Ace.”) They explained that they have a policy about not refunding any orders after shipping labels have been printed. I explained that I, too, have a policy: Do not take money from me without asking. Having involuntarily adjusted my personal policy in my dealings with FitRx/Zealthy, perhaps they, too, could make an exception? No dice. I called the shipper, FedEx, which said that it was completely within the provider’s capacity to cancel a shipment before it had been mailed out. I refused the delivery of the vials and had them returned to the sender, an intermediary pharmacy in Texas. The process was mind-breakingly frustrating, seemingly by design.
…“The obese world is a white-collar crook’s dream,” alleges Sarah Harris, another former customer who claims she was bilked for more than $1,500. She turned to Zealthy back in 2024, when her doctor refused to prescribe semaglutide to her because the treatment was, in Harris’s words, “still newish.” Her insurance wouldn’t cover any weight-loss drugs. So, if she wanted to try them, she’d have to pay out of pocket.
…Ali Garrison, a social media weight-loss influencer who operates the YouTube channel FitFlavorFun, has fallen down what she calls “the Zealthy rabbit hole.” There were Reddit threads and whole Facebook groups collecting stories of consumers who claimed to have been ripped off by the company. “One thing I tell people is to do a Google search,” she advises. “Company Name scams … Company Name reviews … I could go on and on about all problems various telehealths have had.”
Caveat emptor, as the old saying goes. And, in my own case, I am definitely guilty of conducting precisely zilch in the way of this sort of preliminary research. I forked over my Mastercard info to a telehealth website because it seemed glossy, professional, and otherwise sufficiently legit. (I have since canceled that credit card, as a precaution.)
«
Unsurprising that there would be scams around these drugs. But in the UK, you can challenge charges with your credit card company, and then it becomes their problem.
unique link to this extract
How the Open Knowledge Format can improve data sharing • Google Cloud Blog
Sam McVeety and Amir Hormati:
»
As foundation models continue to improve, the lack of relevant context often limits what they can do, especially as they are used to build agentic systems. While these models can help you write code, summarize documents, or analyze a dataset, they still need the right information to produce accurate and actionable results.
That’s why today, we’re introducing the Open Knowledge Format (OKF), an open specification that formalizes the LLM-wiki pattern into a portable, interoperable format. This is a vendor-neutral, agent- and human-friendly standard for representing the metadata, context, and curated knowledge that modern AI systems need.
As published, OKF v0.1 represents knowledge as a directory of Markdown files with YAML frontmatter, with a small set of agreed-upon conventions that let wikis written by different producers be consumed by different agents without translation.
That’s it. No complex compression scheme, no new runtime, no required SDK. A bundle of OKF documents is:
• Just Markdown — readable in any editor, renderable on GitHub, indexable by any search tool
• Just files — shippable as a tarball, hostable in any git repo, mountable on any filesystem
• Just YAML frontmatter — for the small set of structured fields that need to be queryable: type, title, description, resource, tags, and timestamp.
«
This sounds very much like the Semantic Web that Tim Berners-Lee was always dreaming of. This time, perhaps it will stick because they need it to stick for AI agents.
unique link to this extract
Solar generates more energy than coal in US for 1st time • ABC News
Julia Jacobo:
»
The amount of solar power generated in the US is continuing to grow despite efforts from the Trump administration to slow down the renewable energy sector, according to two reports released this week.
The US has generated more power from solar compared to coal for the first time, according to a report by Ember, a think tank focused on the clean energy transition. In May 2026, solar supplied 12.8% of US electricity, while coal supplied 12.2%, according to an analysis of official monthly and preliminary hourly generation data.
A record 45.5 terawatt-hours (TWh) of solar energy was generated in May 2026, exceeding the output from May 2025 by 17%, the think tank found. The record could be broken again in the upcoming summer months, as solar output typically peaks in June and July.
The amount of energy from coal generated in the US has been nearly cut in half in the last five years, falling from 19.7% of total power generated in May 2021 to 12.2% last month. Production of coal power rose slightly in May 2026, to 43.4 Twh, but it remained 11% below May 2025 levels.
«
May of course isn’t the sunniest month. Coal is gradually on the way out.
unique link to this extract
The ‘new joule order’ is here. The west is last to realise • Financial Times
Jeffrey Currie is a senior adviser at Carlyle Group and executive co-chairman of Abaxx Markets:
»
The world has underpriced physical energy security for a decade and the repricing in markets from that complacency was not going to be linear. The Strait of Hormuz crisis is the first major stress test of the new regime. It has revealed which countries prepared, and which are improvising.
China prepared. While Washington and Brussels spent 25 years polarised between green and brown energy, Beijing built both — 1.2 terawatts of solar, the world’s largest nuclear pipeline, an EV fleet now displacing more than one million barrels a day, according to Rhodium group estimates.
None of it was justified as climate policy. It was insurance. An electron can be sourced from coal, gas, sun, wind, or uranium; a combustion engine is married to a single fuel that must cross someone else’s chokepoint. Electrification is the purchase of optionality, and China bought more of it than any nation in history.
That option is now being exercised, and the market is misreading it as weakness. The consensus explanation for falling Chinese energy imports is “demand destruction” — the world’s largest importer buckling. The reality is the opposite. Faced with fuel prices up to 20%, Chinese drivers charge instead of fill: highway EV charging surged 56% year on year over the May Day holiday. Industrial users have switched to domestic coal. In aggregate, China can flex perhaps 2mn barrels a day of demand at will — flexibility no other large economy can approach. What looks like a collapsing customer is a customer that built an exit and is using it.
Environmentalists need not take alarm that some of that flexibility runs on coal. Coal is the bridge, not the destination. China’s renewable and nuclear capacity carries little marginal cost once built: the fixed costs are sunk, the fuel is free, and every additional gigawatt displaces a barrel at little incremental expense. It is the same logic that made asset-light technology businesses so powerful — once the infrastructure exists, the marginal unit costs next to nothing.
The west, meanwhile, is running the old playbook at its physical limit. US crude exports have surged 2mn barrels a day since the war began. Washington calls it energy dominance. Look closer and it is something else: America is exporting its insurance policy.
«
The tanks in Cushing, Oklahoma, are hitting bottom. The oil market is about to hit a tipping point • CNN Business
David Goldman:
»
Cushing, Oklahoma, dubs itself the pipeline crossroads of the world. The tagline is emblazoned on a giant roadside sign fashioned out of pipes on the corner of Main Street and South Stiles Road. It has a valve and everything.
But it’s not just a slogan. In 1912, Tom Slick (his real name) was passing through what’s now Drumright, Oklahoma, when he smelled oil. He bought the land for $1 an acre and started digging, uncovering what was then Oklahoma’s biggest oil well.
Today, neighboring Cushing is the hub of America’s energy market. It literally provides the oil plumbing for the United States. It’s where America’s benchmark West Texas Intermediate oil is priced and warehoused. From there, it’s piped to refineries around the country.
In normal times, Cushing stores around 40 million barrels of oil with capacity of up to 75 million.
These are not normal times. Cushing’s current inventory is 21.6 million barrels, according to the US Energy Information Administration. That’s dangerously close to operational stress levels, the tipping point at which Cushing struggles to supply all of its customers with the oil they demand.
When Cushing’s reserves get below 20 million, they effectively hit empty, scraping the bottom of the barrel of what is largely unusable sludge. And when Cushing runs empty, strange things happen to the oil market.
Unless the Strait of Hormuz opens soon – very soon – we’re probably just weeks away from finding out what that looks like.
…US diesel stocks recently hit their lowest level since 2003. Gas inventories have been falling, too – about 5% below where they were a year ago. Other US commercial crude storage facilities outside of Cushing are also getting drained fast – by 7.2 million barrels last week alone.
«
Not to worry! According to Axios, Trump is on the verge of signing a deal with Iran within the next few hours. Well, perhaps days. Who knows.
unique link to this extract
Brit workers waste nearly six hours a week ‘botsitting’ • The Register
Dan Robinson:
»
Almost all UK workers now have to deal with AI, but few firms report big productivity gains because of all the time lost in hand-holding the systems and cleaning up their mistakes.
So says a report by the Work AI Institute, a research arm of AI biz Glean Technologies.
It claims there are productivity gains to be had from introducing AI-based tools, yet much of this is being negated by the amount of time employees waste making them work – a phenomenon it has christened “botsitting.”
The organization surveyed 1,500 digital workers for “The Work AI Index: UK 2026” report, finding 90% are now required to use AI in their roles, 80% use multiple AI tools every week, and 39% use four or more.
The workers indicate AI automation saves them roughly 12 hours a week, or just under a third of their working week. Yet only 18% agree AI has significantly improved their organization’s performance.
The time freed up isn’t flowing into productive work, it’s being absorbed by the unglamorous human labour required to keep those systems running, according to the Work AI Institute.
For every hour a UK staffer spends getting output from their AI tools, they spend roughly another hour making it usable.
«
It’s still early days, though one has to feel that this in the future the botsitting will just be more complex: they’ll do more difficult things but that will require more careful evaluation.
unique link to this extract
I feel the need, the need for (less) speed • Status-Q
Quentin Stafford-Fraser:
»
When we started towing a boat behind our car, and were consequently limited to 60mph rather than 70mph on major roads in the UK, I found I rather liked travelling that way. Yes, it took a bit longer, but it was more relaxing.
And I’ve noticed that, whether due to a growth in wisdom or to a decline in testosterone — I prefer to think it’s the former — I now tend to drive rather more slowly than I did a decade or two ago. Rose suggests I may just be subconsciously aware of slower reaction times…
But this has led me to propose Quentin’s Law of Optimal Velocity, which is the maximum speed in miles per hour at which you like to drive, and is given by:
Vmax = 120 - agebut I freely admit that this is based on a rather small sample size (errm… one, to be precise) so would be grateful for more data.
«
It does ring true, to be honest. Input welcome.
unique link to this extract
‘Tell him he’s a piece of shit’: Meta’s new AI unit is a total mess • WIRED
Paresh Dave on discontent inside the Applied AI unit inside Meta, where the staff view themselves as doing drudgery:
»
Zuckerberg’s memo also addressed the allegedly dismal situation in Applied AI directly, referring to the unit by its acronym. He suggested the team was a waypoint, not a destination. “Work like AAI is critical to advancing our models and it lets very talented people contribute to those efforts while we create other roles they can contribute to around Meta over the coming months as well,” he wrote.
Engineers selected for the unit have no choice but to join or leave the company, an unusual requirement for highly valued technical employees in Silicon Valley. That’s led some members of Applied AI to describe themselves as “draftees.”
The organization has grown in batches since early April. “It’s crazy to watch people experience the shock of it as each wave comes in,” an early member of Applied AI says.
Some employees are being asked to finish two tasks per week. These involve generating complex software coding problems to help AI scientists better train and evaluate the performance of the latest frontier models. Some of the work is meant to help develop AI agents that generate software or other outputs.
One worker describes the assignment as “mechanical and not creative” and certainly “not using their full skillset and knowledge.” They feel they were hired to develop social media apps for billions of people, but now find themselves assembling data for hundreds of AI scientists to feed to computer chips.
Meta released pioneering open-weights AI models three years ago, but has had mixed results with subsequent releases. Applied AI is among several expensive initiatives Zuckerberg has spun up in hopes that the company can better compete in the growing market for AI services.
Zuckerberg noted in his memo that, unlike some other AI labs, “automating work” was not Meta’s primary focus. “The products we’ll build will range from much more personalized Instagram and Facebook experiences and glasses that help you throughout the day to better tools for small businesses to thrive and create jobs, and personal superintelligence agents that understand your goals and work 24/7 on your behalf to help in the ways you want,” he wrote.
«
| • 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 