Fridges in hundreds of French supermarkets were taken over by hackers who raised their temperature to spoil food. But who was behind it? CC-licensed photo by duncan cumming 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. Warming 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.
AI models may be capable of self-replicating without access to a datacentre • Anthropic
William Zhou and Daniel Freeman, who work for Anthropic:
»
Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial IoT devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off – though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own. In our evals, a 9bn model got inference running on 8 of the 9 hardware types we tested.
Models of 30bn parameters and under can already complete many of the individual steps for replication today, though they make mistakes when composing them end to end. In our simplified evals, GLM-5.2 and stronger models are capable of end to end replication. If small models improve at their historical rate, they’ll match this capability level by the end of this year. Small weight-based replicators may be possible soon after.
Ever-improving AI capabilities make a longstanding concern increasingly realistic: autonomous replication and adaptation (ARA), which involves a model deploying and propagating itself without human control.
In cyber, a worm is a program that copies itself from one computer to the next without human input. Prior work has already shown that AI agents can act like one: scanning their network, executing an exploit, and copying model weights to a target machine across a GPU cluster. Such an “AI worm” could carry its model weights with it as it reproduces (or re-download them from a public hub), and then run inference directly on compromised hardware. Effectively, the weights would become part of the malware payload.
Datacentre GPUs would be a clear, high-value target for such malware. But expensive accelerators aren’t strictly necessary. In this blog post, we show that a distributed botnet of IoT devices—which are cheap and typically poorly defended—could also provide enough compute to run a capable model.
«
This is bad. Like, really, really bad. If you got an AI that was also a worm that could also infect an IoT device, well, there are a zillion of them out there, and they have rubbish security for the most part, and part of any LLM’s training includes how to hack into such devices.
It’s hard to think of what a parallel with the Manhattan Project would be, but it must be possible.
unique link to this extract
Phishing emails are baiting your AI assistant • Malwarebytes
Pieter Arntz:
»
You ignored the suspicious email. You didn’t open the attachment, click the link, or reply to the sender.
Then you asked your AI assistant to summarize your inbox.
Could the message you wisely avoided find its way back onto your to-do list—only this time presented as an urgent recommendation from a tool you trust?
That’s the risk behind phishing emails containing hidden instructions aimed at AI assistants. Instead of relying only on persuading you, attackers are also trying to influence the software that reads your messages. Recent research from Barracuda describes emails designed to target both audiences at once.
For the human recipient, there was a password-protected attachment, with the password supplied in the email. For an AI assistant processing the message, there were hidden instructions intended to influence how it summarized or handled the content.
By including a prompt injection, attackers give themselves another way to reach you. The malicious instructions could try to make the assistant describe the email as legitimate or urgent, encouraging you to revisit it.
In another example, a hidden block instructed a summarizing AI to add a fake priority action: change a vendor’s payment details. That could result in an employee sending money to an attacker’s account.
The attacker isn’t simply asking the AI to repeat a fraudulent request. They’re trying to turn that request into the assistant’s recommendation.
«
Once it was auto-executing Word attachments; now, embedded prompt injections.
unique link to this extract
Broadcom CEO likens many AI investors to casino gamblers • Bloomberg via Yahoo Finance
Gao Yuan:
»
The AI boom has attracted investors whose strategies resemble those of gamblers, Broadcom chief executive officer Hock Tan said, suggesting some of them have little understanding of what they’re betting on.
Tan, one of the chip industry’s most prominent figures, on Monday shared his frank views on the AI industry and the uncertainty that accompanied its spectacular rise. The head of Broadcom, which designs some of the components integral to the global data center buildout, also criticized whether businesses were wasting money on AI initiatives and questioned the longer-term economics of money-losing AI labs.
But he homed in in particular on the rush to invest in a technology many improperly understood — particularly when it comes to large language models on the so-called technology frontier.
“A lot of investors have no clue what they are investing in,” the head of the American chipmaker said at an event in Singapore. They “set very extreme expectations only to find out that it’s almost like they are in a casino.”
With AI companies such as ChatGPT-maker OpenAI reporting billions of dollars in losses, the industry is yet to find a sustainable business model for services whose production requires massive investments in advanced chips, data centers and model development. Tan said that OpenAI reaching annualized revenue of $50bn in September points to a viable business model but offers little guarantee on long-term profitability.
«
In fairness, the same “well I don’t see how this is ever going to be profitable” line was frequently heard early in the dot-com boom. Nobody thought Netscape could make a profit (they were correct), nor Amazon (wrong), nor Pets.com (correct), but they were mightily impressed by WorldCom (fraud) – which is unrelated to Broadcom.
unique link to this extract
Palantir CTO: AI researchers smoke, betting AI will cure cancer • Business Insider
Thibault Spirlet:
»
Forget sleepmaxxing and cold plunging: smoking is the new lifestyle flex among AI researchers.
At least, that’s according to Palantir CTO Shyam Sankar, who said some AI researchers are taking up smoking to signal their belief that AI will eventually cure cancer.
“So many AI researchers have started smoking as a way of signaling to other researchers how much they believe in AI,” Sankar said in an interview on CBS News’ “Face the Nation” on Sunday.
The logic, he said, is: “I believe so much in AI that it’s going to cure cancer. So I’m showing you my chips are all in.”
Sankar did not name any researchers or provide evidence for the claim. Smoking remains a major cause of preventable illness and death, and no AI breakthrough has eliminated the health risks associated with cigarettes.
But the Palantir executive’s remarks echo a countercurrent already visible in parts of the tech industry.
As Business Insider’s Henry Chandonnet reported last month, a small group of founders, engineers, and investors has embraced cigarettes in a sector otherwise known for optimizing their health with cold plunges, sleep tracking, and peptide treatments.
For some of them, a cigarette break is an opportunity to meet people and network. For others, it’s a way to come across as more relaxed and less polished than the typical startup founder.
«
Alternatively, it’s a way to come across as really stupid about your health. It’s like the reverse of Pascal’s Wager: rather than leading a good life because it would turn out badly if there actually were a God and you were judged, you lead a bad life because you believe there is a god in the machine which will cure you.
unique link to this extract
Fridges in French supermarkets hit by cyber attack • BBC News
Ian Aikman and Hugh Schofield:
»
Hundreds of French supermarkets have been hit by a cyber attack that caused their fridges and freezers to heat up, spoiling fresh and frozen food estimated to be worth millions of euros.
Hackers gained control of the software that regulates the temperatures of the refrigeration units, forcing some to warm to 50C (120F) overnight, according to one supermarket chain affected.
The chief executive of Coopérative U said several of its supermarkets had been hit by the “extremely serious and unprecedented” attack.
More than 100 of the chain’s supermarkets were targeted, while 180 stores from Intermarché and its Netto brand were affected, with a spokesman saying that another 100 shops were urgently disconnected from the internet to prevent the same thing happening to them.
Other major food retailers including Carrefour, Monoprix and E.Leclerc were also hit by the attack, which targeted software provided by Danish engineering firm Danfoss.
The software allows retailers to manage their cold units remotely.
Normally, supermarket freezers go through a short period of defrosting overnight, in which temperatures rise slightly. But the hackers ordered the units to heat up and stay hot, although Danfoss did not confirm any reports about just how warm the fridges and freezers could become in that process.
…The full scale of the attack is unclear, but figures supplied by individual chains suggest about 400 stores were affected out of 15,000 or so overall in France.
Carrefour said fridges in 80 of its 6,000 stores were hacked, while Monoprix said 5% of its 400-plus stores had been hit, with food removed from sale if its condition could not be guaranteed.
«
Retail giant Ahold Delhaize takes systems offline after cyberattack • Bleeping Computer
Sergiu Gatlan:
»
Retail giant Ahold Delhaize, one of the world’s largest food retail chains, confirmed today that it was forced to take some systems offline after a “cybersecurity incident.”
As a wholesale company and multinational retailer, Ahold Delhaize operates more than 9,400 local stores across Europe, the United States, and Indonesia, employs nearly 400,000 people, and serves around 60 million customers each week through physical retail locations and online.
Ahold Delhaize reported yearly net sales of over $104bn in 2024, and it owns a wide range of brands, including Delhaize, Maxi, Mega Image, Albert, bol, Alfa Beta, Etos, Gall & Gall, and Profi in Europe, and Food Lion, Stop & Shop, Giant Food, and Hannaford in the American market.
In a succinct press release published earlier today, the company added that it’s working to mitigate the incident and that its stores are open to customers.
“Ahold Delhaize is experiencing a security issue within its network. Our teams are taking steps to assess and mitigate the issue. This includes taking some systems offline to help protect them. At this time, our stores are open and serving customers,” it said.
“Protecting our infrastructure in the best interest of our customers, associates and partners is a top priority. We apologize for any inconvenience this issue may have caused customers and partners.”
However, employees at the Stop & Shop American regional supermarket chain are having trouble accessing their schedules and pay stubs, while customers have been unable to use websites, apps, and customer service platforms at Albert Heijn, Etos, and Gall & Gall stores in the Netherlands, and Delhaize in Belgium.
«
One might wonder if the attacks on the French fridges and now this Dutch company are related. Too soon to know but worth bearing in mind.
unique link to this extract
Epic Games reportedly delays Star Wars and Marvel shooter, as the Fortnite maker embraces AI-accelerated development • MSN
Kaan Serin:
»
Epic Games has reportedly turned to generative AI to finish up development on a Star Wars and Marvel crossover shooter, which has been internally delayed.
That’s according to a Bloomberg report from journalist Jason Schreier, citing anonymous sources familiar with the unannounced game. Epic’s in-development project – a collaboration with Disney – is said to have hit a speed bump, and the Fortnite maker is apparently telling its developers to embrace more gen-AI tools for the rest of the game’s production.
The new shooter purportedly features characters from both the Marvel and Star Wars series, and the company expected to have it out the door in March 2027. Now that the crossover project has slipped, Epic reportedly told its staff to adopt an “AI first” production pipeline and finish the game with tech such as Anthropic’s Claude.
In an internal message seen by Bloomberg, one game director told developers there would be “no more hand-written code” and art work should be “generated with full AI workflows” – much to the surprise of many workers.
“We’re embracing AI-accelerated development, led by Epic developers who are the visionaries and creative drivers for everything we do,” Epic’s communications director, Jake Jones, said in a statement to Bloomberg.
«
That’s quite a Rubicon – “no more handwritten code”.
unique link to this extract
Meta AI data centre workers drive business boom in rural Louisiana • Business Insider
Lauren Edmonds:
»
The construction workers building Meta’s largest-ever data center are arriving in a tiny Louisiana town with more than specialized skills and manpower. They’ve also got cash to spend.
It’s a fact that local entrepreneurs are leveraging to build new businesses. “It’s been a good opportunity for everybody,” David Rodgers, the cofounder of Ice Queen, a local business, told Business Insider. “I tell people, you can’t stop it unless you have $50 billion. You might as well get on board and go with the flow.”
Meta said the data centre, dubbed Hyperion, will span nearly 10 million square feet in rural Holly Ridge and will support more than 7,500 jobs at peak construction.
That’s a workforce larger than the population of Aspen, Colorado. While the influx of workers is straining the local housing market and disrupting aspects of daily life, their presence has also boosted the region’s economy.
…In June, a study commissioned by Grow NELA said that consumer spending and capital investments made during the data center’s construction period will generate $162.9m in sales tax revenues at the local and state levels.
While Meta has temporarily transformed Richland Parish into a modern-day boomtown, most of the construction workers will eventually move on to other jobs, taking their spending with them. It’s a reality that business owners in Richland Parish know is on the horizon.
[Restaurant owner Tim] Allen said Holy Tacos would likely continue catering for other businesses in the area and selling out of their restaurant. TJ and Kaycie said HeBrews Coffee & Eats will lean on the clientele it built before Meta arrived. The Crawfords [who do house rentals] said they are considering either selling the properties or continuing to rent them after construction ends. Ice Queen has diversified and plans to rely on other large-scale clients in the area.
Louisiana Economic Development said Meta is expected to work on the Richland Parish site through 2030.
What happens after that is anyone’s guess. For now, though, there’s money to be made.
«
Global smartphone shipments hit a new low with 20% yoy decline, lowest in 14 years • Counterpoint Research
»
Global smartphone shipments declined 20% YoY in Q3 2026, the lowest Q3 level since 2012, according to preliminary estimates from Counterpoint Research’s Market Monitor service. The market decline was close to Counterpoint’s expectations, as relentless increases in DRAM and NAND prices intensified memory cost shock in Q3. This sharply inflated BOM (Bill of Materials) costs and deepened supply shortages, significantly disrupting OEM shipments. In response, OEMs passed on higher component costs to consumers, driving up global smartphone retail prices by 20% so far in 2026 and further increasing market average selling price (ASPs).
Commenting on the global smartphone market, Senior Analyst Shilpi Jain said, “Q3 was the second successive quarter to take the full brunt of the memory crisis. The extreme memory shortage is pushing the smartphone market into a more selective phase, reshaping OEM strategies and even consumer behavior. Consumers are holding on to their devices longer than estimated, increasingly considering refurbished devices and prioritizing higher-value devices, while OEMs are adjusting portfolios, pricing, and market strategies to protect profitability. This is creating greater pressure across the mass market, while premium demand remains comparatively resilient, widening the gap between premium and price-sensitive segments.”
«
Same decline as the PC market, which suggests that price has a uniform effect on buyers (ie price elastic).
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
