Information Technology · Friday, 18 September 2026

01 Briefing what happened

The EU wants no social media under 13 and one hour a day at 13 and 14. First, every app must find out who is a child.

Information Technology 46 sources

The European Commission proposed the EU Kids Act on Thursday: age checks on every new account, no addictive design for anyone under 18, and chatbots off by default for children. It now needs the European Parliament and 27 governments to agree.

the youngest age at which a child could use social media, and only through a parent-run account

from 3 to 12, children could use only child-friendly video services [2]

1 hour

the daily limit for children aged 3 to 14 under the proposal

for 13 and 14 year olds it runs through a parent-controlled 'mini account' [2]

of a company's worldwide yearly sales, the largest possible fine

fast-track investigations must finish within 90 days [2]

92%

of Europeans in an EU survey who called stronger online protection for children a top priority

the Commission cites it as the case for the law [2]

The lead story — what happened

The European Commission, which writes the European Union's draft laws, proposed the EU Kids Act on Thursday. [1]
Children under 13 would not be allowed on social media. From 3 to 12 they could use only child-friendly video services, through a parent's account, for one hour a day. [2]
At 13 and 14, a child could have a 'mini account' run by a parent, with few contacts and one hour a day. At 15 they could open their own account. [2][3]
For everyone under 18, apps would have to switch off infinite scroll, reward tricks, night-time notifications, feeds built from tracking, and messages from strangers. [2]
AI chatbots and AI companions would be off by default for children, and must not be built to make a child emotionally dependent on them. [1][2]
Apps and app stores must check a user's age when a new account is opened. Existing accounts must be checked for under-15s within six months of the rules starting. [2]
Companies that break the rules could be fined up to 6% of their worldwide yearly sales, and must pay a fee that funds the regulators watching them. [1][4]
The age limit was argued over until Monday. The Commission's expert panel wanted 13, France wanted 15, and the European Parliament had asked for 16. [5][2]
France's own under-15 ban was struck down in August by its Constitutional Council, its top constitutional court, saying it was too harsh and did not protect users' data well enough. President Emmanuel Macron then asked for one rule for the whole EU. [6]
CCIA Europe, a lobby group whose members include Google and Meta, said collecting age and identity data at this scale would attract criminals. [1]
The Commission's own age-check app is meant to prove a person's age without revealing who they are. Researchers got round it in under two minutes. [5][6]
It is only a proposal. The European Parliament and the member governments must agree a final text, and Estonia and Belgium oppose age bans outright. [2][5]

Who is involved

Ursula von der Leyen

president of the European Commission, the EU's executive arm; she announced the age limits in a speech on Wednesday [3]
Henna Virkkunen

the EU's tech chief; she presented the act with von der Leyen [4]
CCIA Europe

a lobby group for tech companies including Google and Meta; it warned the age data would be a target for criminals [1]
France

an EU member that pushed for 15; its own under-15 ban was struck down by its top constitutional court in August [5][6]
Estonia and Belgium

EU members that oppose age-based bans and want the platforms regulated instead [5]

How it unfolded

March Von der Leyen sets up a panel of more than 60 experts on children online [2]
August France's top constitutional court strikes down France's under-15 ban [6]
Mon 14 Sep A draft seen by Reuters shows a plan restricting under-15s [4]
Wed 16 Sep Von der Leyen: no social media under 13, no personal account under 15 [3][6]
Thu 17 Sep The Commission presents the EU Kids Act [1]
Next The European Parliament and member governments negotiate a final text [2]

Where this points

The next test is whether the European Parliament, which asked for 16, and governments like Estonia, which oppose any age ban, can agree one number with the Commission. [2][5]

What is pushing on the whole day

The bar and the word are our reading of how hard each one is pushing today. The arrow is where it is heading. The evidence is in the stories below.

Rules for children online Building↑

The EU proposed keeping children under 13 off social media and switching chatbots off for them. [1] Meta changed its AI assistant after it suggested questions about a mother's young daughters. [18]

Cameras and faces nobody agreed to High↑

A woman found herself posted on Instagram after a date wore Meta's camera glasses. [9] A lawsuit says Meta built face data from people's Facebook and Instagram photos. [14] Meta is reported to be preparing glasses with no camera at all. [7]

Personal data reused or leaked High↑

Revolut, a large online bank, sent customers' passports to scammers posing as a government agency. [35] OpenAI pays contractors to read real ChatGPT conversations. [16] Apple will let apps in the EU ask a softer question about tracking. [17]

Money for AI buildings and chips High↑

Anthropic, maker of the Claude AI models, agreed to lease a 2.16-gigawatt data-centre site in Australia, people familiar told Reuters. [20] Amazon took the right to buy up to $340m of Generac, which will supply its generators. [28] Investors put $90bn into hardware start-ups so far this year. [31]

Anger at data-centre costs Building↑

The US House passed a bill letting states make big data centres pay for their own new power lines. [37] Comments against data centres outnumbered supporters nearly four to one under Donald Trump's posts. [38]

The rest of the day

33 more stories on this beat.

Each with its own sources. None of these is a link to the story above.

02
Meta readies glasses with no camera

Meta is preparing smart glasses with no camera, codenamed Luna, to show at its Connect conference on 23 and 24 September, the tech news site The Information reported. [7][8] They would carry six microphones and speakers so the wearer can talk to Meta's AI, and could ship in October. [7] Meta's camera glasses have been banned from New York courtrooms and Britain's largest pub chain. [8] A slimmer headset, Project Phoenix, is expected in 2027. [8][10]

Why it matters — The camera is what drew the backlash, so Meta's reported answer is a pair without one. [8] Microphones in frames that look like ordinary glasses may raise their own worries. [8]
03
Filmed on a date without asking

WIRED spoke to women who were filmed by dates wearing Meta's camera glasses, including one who found the footage on the man's Instagram the next day. [9] Meta sold more than 7 million pairs in 2025. [9] A small light shows when the glasses record, but it is the only warning and people have found ways round it. [9] A California bill that passed both chambers would require visible recording lights that cannot be switched off. [9]

Why it matters — The person being filmed never agreed to anything, and the only protection is a light they may not notice. [9] Australia's online safety regulator has asked makers to blur faces automatically. [9]
04
Meta sued over face data from photos

A proposed class action says Meta turned faces in Facebook and Instagram photos into faceprints, a kind of biometric ID, without consent. [14] It follows a WIRED report that code for a face-recognition feature called NameTag sat inside Meta's glasses app, downloaded more than 50 million times. [14] The plaintiffs include a California man and his 10-year-old daughter. [14] Illinois law allows $5,000 for each intentional violation. [14]

Why it matters — Meta paid $650m in 2020 and $1.4bn to Texas in 2024 over earlier face data cases. [14] Meta calls the new suit without merit and says nothing from NameTag has shipped. [9]
05
Contractors read real ChatGPT chats

OpenAI pays hundreds of contractors to read real ChatGPT conversations and rate the replies, under a project called Lily, the news site 404 Media reported. [16] Reviewers do not see names, but they can see a summary of what the user has asked before. [16] A filter strips personal details first, and OpenAI's own page says it can miss some. [16] The setting that allows this is on by default for free and paying consumer accounts, and off for business accounts. [16]

Why it matters — People who assumed a chat was private may have had it read by a stranger. [16] Italy's privacy regulator has already fined OpenAI 15 million euros over how it handles personal data. [16]
06
Revolut sends passports to scammers

Revolut, a London-based online bank with more than 80 million customers, confirmed that scammers got customers' data by sending requests from a real government agency's email address. [35] The data included passports, driving licences, birth dates, addresses and possibly verification selfies and statements. [35][36] Revolut says a 'limited' number were affected and will not name the agency or the countries. [35] It blocked the address and told police and regulators. [35]

Why it matters — A passport copy cannot be changed like a password, so the risk to those customers lasts. [36] Banks are trained to answer official requests quickly, which is exactly the habit this attack used. [36]
07
Apple softens its tracking question in Europe

Apple will let apps in the EU show a gentler version of App Tracking Transparency, the pop-up that asks if an app may follow you across other apps. [17] The new screen drops the word 'track' and offers 'Allow' and 'Reject'. [17] It follows agreements with some European competition regulators, and only the new version will be used in Germany, France, Italy, Poland and Romania. [17] Apps may ask again after a year. [17]

Why it matters — Most people answer the old question by refusing, which cuts income for ad-funded apps like Meta's. [17] Apple says the screens protect privacy, while its own apps can still show personalised ads from data Apple collects. [17]
08
German court makes Meta pay for fake ads

A German court ruled that Meta is liable for fake investment ads that strangers placed on Instagram and Facebook, and ordered it to remove them and pay damages. [15] The case was brought by a German financial website whose logo and founder's image were used without consent. [15] It reported nearly 260 such ads in August 2024, and Meta took up to 62 days to remove some. [15]

Why it matters — The court said Meta cannot claim it did not know, because its own algorithms choose what people see. [15] The ruling is not final, and Meta says it is considering next steps. [15]
09
Meta's AI asked about a mother's daughters

Meta changed the questions its AI assistant suggests after a viral video from Kalie Robins, an Instagram user. [18] Under a clip of her singing in the car with her child, Meta AI suggested asking 'Who is the child passenger?'. [18] It then gathered details and photos of her daughters from her posts and her relatives' posts, including one she says she deleted. [18] Meta said the feature 'missed the mark' and has been fixed. [18]

Why it matters — An assistant built into Facebook, Instagram and WhatsApp pulled scattered family posts into a profile of two children. [18] Meta had already pulled an AI image feature in July after a backlash. [18]
10
Snap puts $2,195 glasses on sale

Snap, the company behind Snapchat, showed what its $2,195 Specs augmented-reality glasses can do at an event in Los Angeles on Wednesday. [12][13] They are a full computer on the face, with displays in both lenses that lay pictures over the real world. [11] Buyers can stream HBO Max, use Spotify and translate speech. [12] The battery lasts about four hours, and a reviewer found the controls finicky. [11]

Why it matters — Snap has beaten Meta to glasses with screens in both eyes, but at a price most people cannot pay. [11][13] Its first reveal, in June, drew mockery, and Snap's shares fell. [12][13]
11
Cohere and Aleph Alpha merge

Cohere, a Canadian AI company, and Aleph Alpha, once called Germany's AI champion, confirmed they will combine under the Cohere name, with head offices in Toronto and Berlin. [19] Schwarz Group, which owns the Lidl supermarkets, will invest 500 million euros and supply computing power. [19] Cohere reported $240m of yearly recurring revenue this year. [19] The deal still needs regulators' approval. [19]

Why it matters — Both sell AI that runs inside a customer's own computers and follows local rules, which governments and big firms are asking for. [19] Mistral, Europe's leading AI model maker, raised 3 billion euros last week. [19]
12
Anthropic leases a site in Australia

Anthropic, the maker of the Claude AI models, has agreed to lease a planned data-centre campus in Australia, people familiar told Reuters. [20] The site is farmland about 250km from Brisbane, planned at 2.16 gigawatts, and would start running in 2027. [20] Its Singapore-based developer, Zerra DC, says it will buy renewable power and pay for its own grid connection. [20] Australia's foreign investment board must approve it. [20]

Why it matters — OpenAI signed for a Sydney data centre last year, and economists expect up to A$150bn of such investment by 2030. [20] Australia's government has promised limits on data-centre use of land, water and power from 2027. [20]
13
How banned Nvidia chips reach China

C4ADS, a research group funded mostly by the US government, traced three routes by which Nvidia's AI chips, banned from sale to China, get there anyway. [21] The largest ran through Megaspeed, a firm based in Singapore whose real owner is unclear, and was worth $4.6bn. [21] Smaller amounts moved through Vietnam, India and Malaysia, and through Chinese universities. [21] Another study estimates about a third of China's AI computing runs on smuggled chips. [21]

Why it matters — The US bans the sales and China discourages the purchases, yet the chips arrive anyway. [21] C4ADS wants more staff for the US office that checks where exported chips end up. [21]
14
Huawei names its next AI chips

Huawei, the Chinese company that makes telecoms gear and AI chips, will launch two new AI chips in 2027, its rotating chairman David Wang said on Thursday. [22] The 960DT is due in the first quarter and the Ascend 960PR in the third. [22] Wang said Huawei has shipped more than 1,000 large AI computing systems to more than 370 customers. [22]

Why it matters — Huawei is the company China relies on to replace the Nvidia chips the US will not sell it. [21][22] Wang said UnifiedBus, Huawei's way of linking many chips to work as one, is key to its next systems. [22]
15
GlobalFoundries lines up two deals

Japan and the US are discussing a chip factory worth 2 to 3 trillion yen, about $13bn to $19bn, to be run by GlobalFoundries, the Nikkei newspaper reported. [23][24] It would come from the $550bn Japan pledged to invest in the US under their tariff deal. [24] Separately, GlobalFoundries and Marvell expanded a deal for chips used in fast optical links inside data centres, made in Vermont. [25]

Why it matters — GlobalFoundries makes older, specialist chips for cars and phones, not AI chips, so the Japanese money would go to everyday parts. [24] Neither government has commented on the plant. [24]
16
MediaTek's first 2-nanometre phone chip

MediaTek, a Taiwanese company that designs chips for Android phones, launched the Dimensity 9600 Pro. [26] It is its first phone chip made on the newest 2-nanometre process of TSMC, the Taiwanese company that makes most of the world's advanced chips. [26] MediaTek sells to Chinese phone makers such as Xiaomi, Oppo and Vivo, and its stock-market value passed that of Qualcomm, the US phone-chip maker, this year. [26]

Why it matters — MediaTek says phone prices are rising as the AI boom strains supplies, and it still expects to win more expensive phones. [26] Nvidia put $3.5bn into a MediaTek bond sale last month. [26]
17
Salesforce raises its long-term target

Salesforce, which sells software that companies use to manage their customers, told investors it expects more than $63bn of revenue in the year to early 2030. [27] Analysts had expected about $59.2bn. [27] At its Dreamforce conference it said up to 1,000 clients have signed up to use its data inside Anthropic's Claude. [27] Its boss, Marc Benioff, said its Anthropic shares will likely be worth tens of billions. [27]

Why it matters — Software sellers' shares fell this year on fears that AI would replace them. [27] Salesforce is answering by selling AI through its own products and by holding a stake in an AI lab. [27]
18
Amazon takes a stake in generators

Amazon got the right to buy up to $340m of shares in Generac, a maker of backup power generators, and Generac's shares jumped more than 40% after hours. [28] Generac will supply generators for Amazon's data centres, with first deliveries worth $2.4bn in 2027 and 2028. [28] Last week Amazon took similar rights to up to $4bn of Qualcomm shares in a chip deal. [28]

Why it matters — Amazon often takes a stake in the suppliers it depends on, from chips to air cargo. [28] Here it is buying backup power for its own data centres. [28]
19
Manus raises at a $4bn value

Manus, an AI start-up founded in China, is raising about $500m at a $4bn value, twice what Meta had agreed to pay for it, people told Bloomberg. [29] China's government forced Meta's $2bn purchase of Manus to be reversed earlier this year. [29] GMI Cloud, which rents out AI computing, is seeking a $300m loan to buy chips for a data centre in Thailand that the Chinese internet giant Tencent will use. [29]

Why it matters — A purchase blocked by China's government left Manus worth twice the price Meta had offered. [29] Lenders now treat the chips themselves as something to lend against, not only the buildings. [29]
20
Factory triples its value to $5bn

Factory, a San Francisco start-up whose AI agents write, test and maintain software for large companies, raised $200m at a $5bn value. [30] That is more than three times its $1.5bn value in April. [30] Backers include the investment firms Blackstone, Khosla Ventures and Sequoia Capital. [30] A rival coding-agent maker, Cognition, raised $2bn at a $48bn value earlier this month. [30]

Why it matters — Writing code is one of the most widely used jobs for AI, and investors keep paying more for the firms that do it. [30]
21
Investors pile into hardware

Matter Venture Partners raised $450m for a second fund that backs chip, robot, energy and quantum start-ups, with ASML, which makes chip-printing machines, and TSMC among its investors. [31] Venture firms put $90bn into this kind of hardware so far this year, up from $23bn in all of 2020, according to the data firm PitchBook. [31] Its founder, Wen Hsieh, spent 17 years at the venture firm Kleiner Perkins. [31]

Why it matters — AI is shaking up software, the business venture firms used to live on, and money is moving to the physical things AI needs. [31]
22
Axelera wins EU supercomputer deals

Axelera, a European chip start-up, will supply chips to EU-backed AI centres in Italy and Luxembourg, working with the computer maker Dell. [32] The EU is turning its supercomputer centres into 'AI factories' so European firms and scientists can get computing power. [32] Axelera makes chips that run AI models rather than train them, and launched a second, larger chip. [32] A Dutch rival, Euclyd, raised 200 million euros. [32]

Why it matters — The EU wants its firms and scientists to have AI computing of their own, to narrow the gap with the US and China. [32]
23
India starts charging for big UPI payments

India's UPI, the phone payment network behind most digital payments in the country, will charge merchants 0.4% on payments above 2,000 rupees, about $21, from 15 October. [33] Buyers still pay nothing, and small merchants are exempt. [33] Payments of 2,000 rupees or less, more than 95% of merchant payments, stay free. [33] The operator says running UPI costs about $2.1bn a year. [33]

Why it matters — India made UPI free for merchants in 2020 to get more people using it. [33] Payment companies such as Paytm and PhonePe will share the new fees. [33]
24
Dangote share sale crashes Nigerian apps

Several Nigerian investing apps went down as people rushed to buy shares in the Dangote oil refinery, Africa's largest ever share sale at $1.6bn. [34] Traffic on Bamboo, one of the biggest apps, rose tenfold within 30 minutes. [34] Cowrywise and InvestNaija also had problems. [34] Nigeria's markets regulator told first-time buyers to use approved channels, and analysts warned of fake sites. [34]

Why it matters — Many Nigerians are buying shares for the first time through their phones. [34] The outage showed how much traffic those apps can take before they fail. [34]
25
US House passes a data-centre power bill

The US House of Representatives passed the Ratepayer Protection Act, with only three members voting against. [37] It sets standards that state regulators may adopt to make data centres using 100 megawatts or more pay for new power plants and lines. [37] The US Senate has not acted. [37] Environmental groups say it is too weak because states can ignore it. [37]

Why it matters — Rising electricity bills near data centres have become an election issue in the US. [37] The bill leaves the actual decision to each state. [37]
26
Trump supporters push back on data centres

The New York Times analysed thousands of replies to Donald Trump's posts praising AI data centres on Truth Social, his social network. [38] Comments against AI and data centres outnumbered those in favour by nearly four to one. [38] Many came from users whose profiles showed them as firm Trump supporters. [38] Several complained about data centres being built in the countryside. [38]

Why it matters — Some of the anger is about sites built in the countryside, where more Republican voters live. [38]
27
Data centres hire away the builders

Data centres are drawing electricians and building crews away from housing, argues Breakingviews, Reuters' opinion column. [39] The US building industry is short about 439,000 workers, and people moving to data-centre jobs get pay rises of 25% to 30%. [39] Research by Ryan Wang at Yale University found that 100 megawatts of new data centre in a US county raises household fixed electricity charges by 0.7%. [39]

Why it matters — The same workers are needed to build homes, which several Western governments are trying to speed up. [39]
28
What the AI build-out must earn

The biggest AI companies will likely spend more than $1 trillion on data centres next year, MIT Technology Review reports. [40] Jessica Wachter, a finance professor at the University of Pennsylvania and former chief economist of the US markets regulator, worked out what they must earn to break even by 2030. [40] Her answer is that their productivity must rise 2.7 times. [40] If it does not, she says, they risk falling behind on debt. [40]

Why it matters — It turns the question 'is AI a bubble?' into a number that can be checked against company earnings over the next few years. [40]
29
Memory shortage settles in for years

Small phone and laptop makers are planning for years of memory shortage, and the boss of Jolla, a Finnish phone maker, expects supply to recover only from 2028. [41] Contract prices for DRAM, the working memory in every device, are rising 13% to 18% this quarter, after 93% to 98% in the first. [41] Memory can be almost 60% of the parts cost of a $400 phone, according to Fairphone's boss. [41] Framework reprices quickly, while Fairphone has not raised prices. [41]

Why it matters — Price rises are slowing but supply is not coming back, so cheap phones may disappear. [41] The research firm Counterpoint forecast phone shipments will fall 13.9% this year, the steepest drop on record. [41]
30
AI security list puts permissions first

OWASP, which publishes a widely used list of top security risks, moved 'excessive agency' from sixth to third in its 2026 list for AI apps. [42] Excessive agency means an AI agent has been given more tools and passwords than it needs. [42] It is the first list partly based on 6,639 real incidents. [42] Since 11 September, the EU's Cyber Resilience Act makes product makers report an actively exploited flaw within 24 hours. [42]

Why it matters — The danger has moved from what an AI says to what it is allowed to do. [42] The EU law covers products, not how a company sets up its agents. [42]
31
Open AI models get a safety project

Base Labs, a research group set up by the AI hosting firm Baseten, launched a safety project with Hugging Face and Goodfire, a company that studies how AI models reach their answers. [43] Open-weight models are ones anyone can download and change. [43] Hugging Face, the biggest site for sharing them, lists more than 6,000 copies with their safety guards removed. [43] The partners have not said how the project will work. [43]

Why it matters — Baseten says safety should be built into how open models are trained and served, not added afterwards. [43]
32
Senate panel backs a free-speech bill

A US Senate committee voted 18 to 10 for a bill letting citizens sue federal officials who pressure companies to suppress lawful speech. [44] Democrat Maria Cantwell pointed to the head of the US broadcasting regulator pressuring TV stations over the late-night host Jimmy Kimmel. [44] Republican Ted Cruz pointed to the Biden administration pressing tech companies. [44] It still needs the full Senate and the House. [44]

Why it matters — Each party accuses the other's government of leaning on tech and media companies. [44] The bill would let citizens take the officials to court. [44]
33
Banking software firm Finastra may be sold

Vista Equity Partners, an investment firm, is exploring a sale of Finastra, which makes payment and lending software used by more than 80% of the world's top banks, sources told Reuters. [45] Finastra could be valued at up to $12bn. [45] The investment firm Blackstone is among the possible bidders. [45] The review is at an early stage and may not end in a deal. [45]

Why it matters — Prices for software firms are weak because nobody knows how AI will change their business, so a sale tests what buyers will pay. [45]
34
UK pressed to write AI rules

A committee of British MPs and peers called for a dedicated AI law and one regulator with real powers. [46] OpenAI said, for the first time according to the news site Politico, that it backs binding UK rules for the few companies building the most powerful AI. [46] The committee estimates police facial recognition scanned about 3 million faces without consent in 2025. [46] Business Secretary Jonathan Reynolds warned against being 'hyperbolic' about AI. [46]

Why it matters — Britain's AI Security Institute tests models but cannot force a company to let it. [46]

02 Lesson why it matters

Data kept for one job ends up doing another

Companies keep passports, chats and photos long after the job they were collected for, and those copies later get read, reused or stolen.

The twist

Handing over a passport or a chat is not one decision made once. The company that keeps it can later use it, show it to others or lose it, without asking again.

How it works

A company collects something for one job: a passport to open an account, a chat to answer a question, a photo to share.
It keeps a copy, because storing is cheap and deleting takes work.
The copy stays, and new uses for it appear: training an AI, human review, face matching.
Or someone outside finds a way to ask for it, as scammers did at Revolut.
The person who handed it over agreed to the first job, not to the later ones.

The same force, elsewhere today

Where this chain is also running, in today's other stories.

Revolut's passports sent to scammers

Passports and selfies were collected to check who customers are, kept on file, and then handed out to someone posing as a government agency.
Contractors reading ChatGPT chats

Conversations were collected to answer questions, kept, and then read by hired reviewers to improve the model.
The Meta face-data lawsuit

Photos were uploaded to share with friends, and the lawsuit says Meta later used them to build face data.
The EU Kids Act

Age checks mean apps must collect data about every user, and a tech lobby group already warns that store will attract criminals.

Where you've seen this

Hospital records

a scan taken for treatment can later be used for research or insurance checks

Shop loyalty cards

a shopping history kept for discounts can be sold to advertisers

Security cameras

footage kept to catch thieves can later be used to follow a person

The catch

The EU's own age app shows a partial answer: it tries to prove an age without keeping an identity, but researchers got round it in under two minutes.

And the whole of it

Anyone who has shown a passport to an app, typed a worry into a chatbot or been caught in a stranger's photo is part of this. None of us can see which of those copies still exist, or who is looking at them today.

03 Truth what's really going on

What is really going on

The European Commission wants every app and app store in the EU to check how old its users are, so that children under 13 stay off social media. That would create a new store of age data about nearly everyone. In the same week Revolut sent customers' passports to scammers, OpenAI was found paying contractors to read ChatGPT chats, and a lawsuit said Meta built face data from people's photos. [1][35][16][14]

Why it works on us — Protecting children is hard to argue against in public, so CCIA Europe, speaking for Google and Meta, did not object to the goal and warned instead that the age data would attract criminals. [1]

Who gains

Ad-funded apps such as Meta's in Germany, France, Italy, Poland and Romania — Apple's softer tracking question drops the word 'track' and offers 'Allow' and 'Reject', and people mostly refused the old one. [17]
Payment companies such as Paytm, Pine Labs and PhonePe — They will share a new 0.4% fee on UPI payments above 2,000 rupees from 15 October. [33]
Salesforce — Its boss says its Anthropic shares will likely be worth tens of billions, and that it will probably sell them to pay off buy-back debt. [27]
Generac — Amazon will buy its data-centre generators, first deliveries worth $2.4bn, and its shares jumped more than 40%. [28]
Manus's backers — After China forced Meta's $2bn purchase to be reversed, Manus is raising money at a $4bn value, twice what Meta had agreed to pay. [29]

Who pays

Every EU user of social media, if the law passes — Apps must check ages when accounts open and estimate the age of existing users, which means collecting more data about everyone. [2][1]
Revolut customers whose documents went to scammers — A passport copy, unlike a password, cannot be changed. [36]
ChatGPT users on free and paying consumer plans — The setting that lets OpenAI use their chats to improve the model is on by default for them, and off for business customers. [16]
People filmed by smart-glasses wearers — A small light is their only warning, and one woman found herself posted on Instagram after a date in New York. [9]
Indian merchants taking larger UPI payments — They must pay the 0.4% fee and are not allowed to pass it on to buyers. [33]
Households near new US data centres — Research at Yale found 100 megawatts of data centre in a county raises household fixed electricity charges by 0.7%. [39]

What nobody knows yet

Open questions from across today’s stories — ours included.

01

Which age limit survives.

The European Parliament asked for 16, France wanted 15, and Estonia and Belgium oppose any age ban, so the final number is up to the negotiation. [2][5]
02

How an app can check age without collecting identity.

The Commission's age app is meant to prove age without revealing who someone is, but researchers got round it in under two minutes. For existing accounts, platforms may estimate age from details such as credit card information. [5][2]
03

How many Revolut customers were affected, and which government agency's email was used.

Revolut says only that the number was 'limited' and will not name the agency, the countries or how long the requests went on. [35][36]
04

Where OpenAI tells ChatGPT users that people may read their chats.

404 Media asked and OpenAI did not answer; after publication it pointed to a help page. [16]
05

Whether Meta's camera-free glasses are real.

The only source is a report by The Information, and Meta has not announced them; its Connect conference is on 23 and 24 September. [7][8]
06

Where the faceprints in Meta's glasses app came from.

The lawsuit says Facebook and Instagram photos, but admits Meta has never disclosed which images, if any, were used. [14]
07

Who really owns Megaspeed, the firm behind $4.6bn of chips reaching China.

C4ADS calls its ownership unresolved, and it is unclear who owns Swiftdata, the Singapore firm that bought it in 2023. [21]
08

Whether the Japan-funded GlobalFoundries plant will happen.

The only report is the Nikkei's, which gave no sources, and neither government has commented. [24]
09

How much India's new UPI fee will raise, and who gets it.

The operator, NPCI, did not say how it reached its $2.1bn cost figure or how the money will be split. [33]

04 Hope carry this

Since 11 September, companies that sell connected products in the European Union must warn their national cyber-response team within 24 hours when a flaw in them is being actively used by attackers.

Also true today

The US House of Representatives passed a bill with only three votes against that lets state regulators make large data centres pay for their own new power plants and power lines, instead of passing the cost to households.
Meta fixed its AI assistant after a mother showed it suggesting questions about her young daughters, days after her video spread.