A second paper suggesting vaping is more dangerous than smoking has been retracted due to errors in analysis. Odd how this keeps happening. CC-licensed photo by Lacey Krusmark 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. Puffed up. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Anthropic releases its first Mythos-class model Claude Fable • The Verge
Robert Hart:
»
Anthropic just announced Claude Fable 5, a new AI model it said is the most powerful model it has ever made widely available.
According to the company, Fable 5 “shows exceptional performance in software engineering, knowledge work, and vision,” with its lead over other models growing as tasks become longer and more complex.
Fable 5 marks the first broad release from Anthropic’s Mythos class of AI models, after the company said the family was so capable at cybersecurity tasks that it was too dangerous to release publicly. Anthropic said the release was “made possible by new safeguards that block responses in specific high-risk areas,” with the system falling back to Claude Opus 4.8 — a model it praised for “honesty” when it launched last month.
Anthropic singled out cybersecurity and biology as two domains where the safeguards may block responses, both areas widely considered sensitive topics for advanced AI systems. The company said that in testing, 95% of Fable sessions ran entirely on Fable responses, without falling back to Opus 4.8.
The company is also releasing Claude Mythos 5, but provided few details on what that means. In a blog, Anthropic said Mythos 5 is the same underlying model as Fable 5, “but with the safeguards lifted in some areas.” For now, access appears limited to the steadily expanding group of organizations granted access to Claude Mythos Preview through Anthropic’s — not entirely watertight — private Project Glasswing initiative. Those users will be able to upgrade to Mythos 5, Anthropic said, adding that it plans to “expand access over time through a more systematic trusted-access program.”
Anthropic did not respond on the record to The Verge’s request for comment explaining how either model relates to Claude Mythos Preview or why the models are numbered “5” when there do not appear to be any previously released Mythos or Fable models.
«
Those (hopeful?) guardrails around cybersecurity and biology are because they don’t want this system helping people hack into systems, or devising (for example) new viruses. Let’s all cross our fingers that those guardrails are really, really good.
It does feel like all these models are cast on the world entirely in beta form; there’s no way they’ve been properly tested against all circumstances, because how could they be? But the ramifications are much bigger than an email client with an occasional bug.
unique link to this extract
Reuters and Time adopt bot-blocking whitelists to rein in AI crawlers • Digiday
Sara Guaglione:
»
Reuters and Time are taking a new approach to bot blocking. Both publishers have recently started blocking all AI bots by default, and created whitelists of approved bots to access content on their sites.
It’s a strategy already adopted by publishers like People Inc. at the beginning of this year and The Atlantic at the end of last year. Reuters and Time made the decision to block bots by default last month.
“We saw that there was an imbalance between the value that publishers like Reuters provide and the value… that Reuters receives in kind, and so instead we went from a default allow-all to a default disallow all,” said Josh London, head of Reuters Professional, who oversees the direct-to-consumer and direct-to-professional businesses. “Our content costs money to create. It has significant value, and the access to it, we feel, must be earned.”
Time now allows about 70 bots to access onsite content, according to its COO Mark Howard. Those bots range from crawlers run by big AI labs and social platforms, to the automated systems Time uses to keep its own website running. Time uses ScalePost to manage AI bots.
“Now we’re starting to think about: as the volume of bot traffic continues to increase significantly, and we see through a number of our vendor partners that we have very high domain authority with AI bot traffic, there’s value in that,” Howard said. That perceived value, he added, can help support the AI visibility product Time is developing to sell GEO insights to brand clients.
Reuters is blocking bots by default using robots.txt files. That method is far from foolproof. It tells web crawlers which URLs they can, but it relies on voluntary compliance — and many AI bots ignore it. A Tollbit report found that 30% of total AI bot scrapes in Q4 2025 did not abide by explicit robots.txt permissions.
«
And yet somehow the Internet Archive can’t do this? Peculiar, isn’t it. (Thanks Gregory B for the link.)
unique link to this extract
Crackdown on tech platforms will go ahead despite US intervention, says No 10 • The Guardian
Dan Milmo and Jessica Elgot:
»
White House displeasure over the prospect of an under-16 social media ban will not deter the UK from cracking down on tech platforms, the British government has said.
The technology secretary, Liz Kendall, told the Guardian she was not concerned “in the slightest” by the Trump administration’s intervention in the debate over restrictions, after the US embassy in London posted a notice warning against a ban.
Kendall added that nine out of 10 respondents to a government poll supported an under-16 ban.
Kendall said she was “very happy to read any submission anybody makes” but her priority was “British young people”. Kendall denied there was any tension between seeking investment into the UK from US AI companies and implementing regulations that affect major American tech firms. “I think companies will continue investing in Britain,” she said. “My focus is on what is right for British parents and British families.”
The government is set to announce some form of social media ban for under-16s next week, alongside other restrictions such as a possible block on conversations with strangers on gaming platforms. Limits on AI chatbot use are also under consideration.
Asked about the Trump administration’s intervention, a Downing Street spokesperson said: “We will always act in the UK’s national interest and protecting young people is no different.”
In a submission to a government consultation on online safety, the US government came out against “prescribed one-size-fits-all government restrictions” and “blunt regulatory instruments” to address online harms to children.
«
The Trump admin going against this is probably the best advert in its favour that Starmer’s government could have wished for.
unique link to this extract
One day after discovery, Meta pulls facial recognition code from its smart glasses • Wired via Ars Technica
Dhruv Mehrotra and Dell Cameron:
»
One day after WIRED revealed that Meta had quietly embedded an unreleased face-recognition system into an app installed on more than 50 million phones, the company removed it, according to a WIRED analysis of the latest version’s code.
The most recent version of Meta AI, a companion app for its line of smart glasses, strips out the unactivated software components that powered the system Meta internally called NameTag. The version published the day of WIRED’s report included several code libraries explicitly named for face recognition. Friday’s release includes none of them.
Andy Stone, Meta’s vice president of communications, told WIRED on Monday that the feature is purely exploratory, adding: “No final decision has been made on what to do here, if anything.”
On Thursday, WIRED reported that Meta had quietly integrated substantial portions of the NameTag system into the Meta AI app. Though never publicly enabled, the feature was designed to convert faces captured by the glasses into unique biometric signatures, commonly known as faceprints, and compare them against a database of faceprints stored on the user’s device. WIRED also found that faces the system failed to recognize were cropped, indexed, and stored locally for future processing.
NameTag first surfaced in February, when The New York Times, citing internal Meta documents, reported that the company was developing face recognition for its smart glasses and weighing a launch as soon as this year. One memo reportedly described releasing it during a “dynamic political environment,” when privacy and civil liberties advocates would be distracted. Last week, WIRED reported that much of NameTag’s machinery was already built into the Meta AI app, downloaded by millions of users, as early as January, even as Meta publicly said it had made no final decision about face recognition.
After WIRED’s report, Stone dismissed the findings, writing that the company couldn’t answer questions about how the system would work because “the feature does not exist.” Andrew Bosworth, Meta’s chief technology officer, called the reporting “incredibly misleading” and “absolutely dishonest.”
«
Just utterly shameless by Stone and Bosworth. The feature did exist; it wasn’t enabled.
unique link to this extract
Meta AI bug allowed hackers to take over Instagram accounts • The New York Times
Mike Isaac and Eli Tan:
»
In March, a group of hackers discovered a bug in a Meta customer service tool that allowed anyone to use an artificial-intelligence-powered chatbot to reset the passwords for Instagram accounts. All the hacker had to do was ask the chatbot to change someone’s password — and it would be done.
Roughly 34,000 Instagram accounts were affected, including the accounts of the home security monitoring company SimpliSafe and a senior official in Mr. Trump’s Space Force department, according to internal Meta documents viewed by The New York Times. In the Space Force official’s case, hackers began posting pro-Iran messages comparing the war in Iran to US involvement in Vietnam in the 1960s.
Of the 34,000 accounts, 20,000 were breached, giving hackers access to the related email addresses, phone numbers, birth dates and other personal data. More than 3,500 of the accounts had their user names taken over and changed from the hack, according to the internal documents. Meta has said it could not determine what information was viewed or stolen by the attackers.
In a statement, Meta said it had fixed the flaw, which was reported by 404 Media this month, and secured the affected accounts.
“Some of our internal back-end checks failed in this instance, but it wasn’t due to the AI agent itself, and we’ve addressed the underlying cause,” said Andy Stone, a Meta spokesman, adding that it was notifying regulators and people whose accounts were affected. The company said that because of its new automated customer service programs called agents, the number of users who were able to recover hacked accounts in the United States and Canada increased by 30% last year.
…The incident was another AI-themed hiccup for Meta as it tries to remake itself using the technology. The company, which also owns Facebook and WhatsApp, is not only integrating AI into its apps but spending billions to keep pace with rivals like Anthropic and OpenAI to develop cutting-edge AI. Mark Zuckerberg, Meta’s chief executive, has said his company’s future depends on quickly shifting to an AI-first organization.
«
Have to wonder if this experience will make them weight that at least a little more carefully. The scale of this attack is small compared to the total number of Instagram accounts, but that’s only because they targeted high-profile ones.
unique link to this extract
Second Patel vape study retracted as lung disease paper is pulled over data concerns • Clearing the Air
Ali Anderson:
»
A second vaping-related NHANES paper involving Urvish Patel has been retracted, intensifying scrutiny of research used to suggest serious health risks from vapes.
The Journal of Investigative Medicine has pulled a 2023 paper titled ‘E-cigarette use and prevalence of lung diseases among the US population: a NHANES survey,’ after concerns were raised about the accuracy of the data behind it.
The paper had examined links between vape use and lung disease using the US National Health and Nutrition Examination Survey, a major public health dataset often used by researchers to study health trends across the American population.
But in a retraction notice published on 27 May, Sage and the journal editor said the article could no longer stand. “Sage was contacted by a reader who raised concerns about the accuracy of the data the article derived from the NHANES survey results,” the notice said.
The journal said its internal investigation also identified concerns around authorship contribution and the number of people on the byline. “The author’s response to our queries did not resolve the concerns we discovered during this investigation,” the retraction notice said. It added: “Due to our inability to verify the author contributions on the byline and the unresolved concerns we have of the accuracy of this study’s data, this article has been retracted.”
“The authors did not respond when notified of this retraction,” it said.
The retracted lung disease paper was authored by Sudha Dirisanala, Srishti Laller, Naga Ganti and others, with Urvish Patel also listed among the authors.
It follows the retraction of another vaping-related NHANES paper involving Patel: Effect Comparison of E-Cigarette and Traditional Smoking and Association with Stroke – A Cross-Sectional Study of NHANES, published in Neurology International in 2022.
That paper had reported that vape users had a higher risk of early stroke than traditional tobacco users. It was later cited in media coverage, featured in an anti-vaping campaign and included in a contested meta-analysis, according to Retraction Watch. MDPI retracted the stroke paper in December 2025 after concerns were raised about major errors in the data analysis.
«
You have to wonder: if Patel just very bad at analysing data, in a way that accidentally keeps suggesting vapes are worse than tobacco, or is there some other motive?
unique link to this extract
Bank of England warns of AI scams as deepfakes of Farage-Bailey fight spread • The Guardian
Dan Milmo:
»
The Bank of England has warned the public against falling for AI-generated scams after deepfake videos of Nigel Farage fighting its governor spread online.
Andrew Bailey, the head of the BoE, said AI-generated content related to central banks was spreading and urged people to be “vigilant”.
He spoke out after the videos of the Reform UK leader and Bailey fighting on the set of BBC One’s Question Time appeared on the social media platform X. The videos showed the men being separated by police officers and even depict Farage holding a gun while he grapples with Bailey.
Bailey urged the public to report the videos so they could be taken down. “Unfortunately, fake adverts impersonating the Bank of England and other central banks are on the rise,” he said. “These scams are designed to criminally exploit the public, especially the vulnerable, when they are online. I would urge everyone to stay vigilant and report these scams. That way authorities can better root out digital deception like this and permanently remove the fraudsters responsible for what is a truly online scourge.”
…The Bank has raised concerns about the posts with Reform UK and with social media platforms.
«
With three weeks of this financial quarter’s 12 weeks remaining, advertising prices have fallen so low on X that crypto scammers were able to stuff these ads into people’s timelines. That demonstrates that advertisers have not returned, and so X becomes an easy mark for these scams (which lure people to pages that have “special time-limited offers” and encourage them to put up some money. It only needs a tiny percentage of those who see the ads to click them, and a tiny number of those to pony up their money, and the scammers have made a profit. By using a network of bot accounts, they make it hard for X to close down the scam.
Musk’s insistence that X would get rid of the bots is hollow, and empirically disproven. Social media is balkanised and broken.
unique link to this extract
Most of our customers’ fraud cases began on Meta, says Lloyds Bank • The Times
Hugo Daniel:
»
More than two thirds of fraud cases reported by Lloyds Bank customers started on Meta platforms.
Those in their late twenties and early thirties are the most at risk from scams on Facebook, Instagram and WhatsApp, according to data collated by the bank.
Concert, festival and sporting event ticket scams are the most prevalent. Under-25s, including children under 18, are most at risk from ticket fraud, according to the bank. Fake tickets for Taylor Swift and Peter Kay gigs, Liverpool Football Club matches and Alton Towers were among the scams reported.
Writing for The Sunday Times, Liz Ziegler, Lloyds’ fraud prevention director, said 68% of fraud reports from their customers started on a Meta platform. “Customers tell us they feel upset, embarrassed and shaken,” Ziegler said. “This is deeply personal and it can take a long time to recover.”
The average claim value submitted to the bank is now above £500, an increase of about £100 from last year.
The bank’s report comes after The Sunday Times revealed at least 260 fraud victims had expressed interest in joining a group action claim — the British equivalent of a class-action lawsuit — to try to recoup their losses from Mark Zuckerberg’s £1.2 trillion company.
«
Hard to know how the collective action claim would proceed – can one really pin direct responsibility for what appears on Meta platforms on Meta? – but it might be interesting to pursue. At least crypto scams seem to have receded, for now.
unique link to this extract
My prodigal brainchild • Graphomane
Neal Stephenson, the SF author who wrote “Snowcrash”, which gave us the idea of the metaverse, on why Meta’s idea of the metaverse failed:
»
When I was working at [augmented reality glasses company] Magic Leap, and people asked me why I thought that was a good idea, I would ask the rhetorical question: “do you really think that twenty years from now everyone is still going to be going around all day staring at little rectangles in their hands?” At the time it seemed obvious to me that the answer was no.
Reader, I have changed my mind. Twenty years from now, everyone is still going to be staring at handheld rectangles. Or at least that is the case if the only alternative is wearing things on their faces. Maybe this should have been obvious to me given the amount of time, effort, and money people put into making their faces look as good as possible.
A possible workaround is to keep refining and miniaturizing the devices to the point where they just look like eyeglasses. This, however, turns out to have the unintended side effect of making these things seem sinister. It happened with Google Glass, which instantaneously spawned the term “glasshole,” and it has happened again with Meta’s product that looks like normal, albeit heavy-framed glasses.
When someone around you is staring at a rectangle in their hand, it might be incredibly annoying, but at least you can tell they’re doing it. When someone’s wearing a head-mounted display, on the other hand, you don’t know whether they are looking at you or not.
Likewise, when someone holds up their phone and aims it at you, it’s obvious that you are on camera. That’s not true in the case of glasses or goggles. So it’s creepy.
…Goggles were the ubiquitous visual signature of Cyberpunk. This, combined with the amount of R & D that has been poured into making various head-mounted displays by tech companies over the last couple of decades, has forged an unbreakable connection in many people’s minds between the Metaverse and goggles.
In 1990, when I was writing Snow Crash, we experienced all computer graphics through massive, heavy CRTs with terrible resolution. The images were flickery and blurry. Rendering pictures of three-dimensional scenes was in its infancy. It seemed entirely reasonable to think that the future would be all about head-mounted displays that could render stereoscopic (simulated three-dimensional) imagery.
This is not actually what happened.
«
Stephenson is acute, and insightful.
unique link to this extract
From the doorstep to the dining room: new DoorDash survey data reveals the full picture of the modern restaurant guest • DoorDash
DoorDash:
»
AI is changing how consumers decide where to eat, and restaurants’ digital footprints are helping shape what gets recommended.
The AI dinner prompt is here: 22% of consumers have used an AI tool like ChatGPT or Google Gemini to help choose a restaurant, making AI a new discovery signal to watch as consumers use these tools to find something new, compare nearby options, and search by cuisine, occasion, or value.
The new rules of restaurant search: According to Yext research, restaurant listing sites like DoorDash account for more than 41% of the sources AI tools cite when recommending restaurants. Operators are already responding by strengthening the basics, like updating menu information (39%), managing reviews (34%), and improving photo quality (32%).
The opportunity doesn’t stop at search: Three-quarters of consumers say they’re comfortable using AI for reservations, but only 28% of operators are using AI to manage calls and customer service. With an estimated 40% of reservation calls going unanswered industry-wide, AI gives restaurants another way to capture demand and keep front-of-house teams focused on the guests in front of them.
«
From DoorDash’s 2026 Restaurant Industry Trends Report. This looks like real consumer adoption of AI, almost surely via Google.
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 