The New York Times and three other papers sue Brave, saying the browser maker copied their articles and sold them to AI firms

The stories

33 stories on this beat today.

Every story told the same way: what happened, who is involved, why it matters, how it works, the truth and the good news.

1. 01

   The New York Times and three other papers sue Brave, saying the browser maker copied their articles and sold them to AI firms

   30 billion

   web pages in Brave's search index, the Times says

   the papers say Brave never named the program that collected them [2]

   5

   AI companies named as buyers of access to that index

   Anthropic, Perplexity, Mistral, xAI and Cohere [1]

   2 million

   articles the 3 Alden papers say are affected

   they want damages, Brave's profits and a jury trial [2]

   95.7%

   less traffic sent back by AI crawlers than by Google search

   a study the papers cite [2]

   What happened

   • The New York Times sued Brave Software on Monday in a federal court in Manhattan. [1]
   • Brave makes a privacy-focused web browser and runs its own search engine, Brave Search. [2]
   • Hours later, the New York Daily News, the Chicago Tribune and the Orlando Sentinel filed a similar suit. [2]
   • Both suits say Brave copied their journalism and sold access to it to AI companies. [2]
   • Brave's search index holds over 30 billion web pages, the Times says. [3]
   • The index is Brave's own copy of billions of web pages, which its search engine looks through. [3]
   • The papers say Brave never named the program that collects those pages. [2] Websites block such programs by name, so they could not block this one. [2]
   • Brave sells access to its index through a service called the Search API. [2]
   • Both suits name 5 buyers: Anthropic, Perplexity, Mistral, xAI and Cohere, all AI companies. [1]
   • Brave's own marketing says the service is used in training and grounding big AI models. [2] Grounding means a model reads fresh pages to answer a user's question. [4]
   • A Brave blog post said Anthropic used it to train Claude 2, an older Claude model. [2]
   • The suits also target a browser setting called Paywall blocker. [2]
   • Brave promotes it as a way to 'Bypass Paywalls', the Times says. [3]
   • One Reddit user said it let them read the Times without an account. [2]
   • Brave's browser blocks the papers' adverts by default, while Brave sells adverts of its own. [2]
   • The papers say this lets Brave earn from their journalism without sharing the money. [2]
   • The 3 Alden papers say more than 2 million of their articles are affected. [2]
   • They claim copyright breaches and getting round technical locks, which a US law called the DMCA forbids. [2]
   • They want damages, Brave's profits and a court order to stop it, before a jury. [2]
   • A Times spokesperson said Brave 'ignored repeated requests by The Times to stop'. [4]
   • 'Buying stolen goods secondhand doesn't make them legitimate,' said Steven Lieberman, the other papers' lawyer. [4]
   • Perplexity, one of the AI firms named, said the Times 'is overpaying its attorneys and underpaying its journalists'. [4]

   Who is involved

   Brave made more than $100m in revenue in 2025 and has about 130 million monthly users, the papers say. [2]

   • The New York Times one of the biggest US newspapers; it filed the first suit and has also sued OpenAI and Perplexity
   • 3 Alden-owned papers second suit
   • Perplexity named as a buyer; it hit back
   • Brave Software the 11-year-old company behind the Brave browser and Brave Search; it is being sued
   • New York Daily News, Chicago Tribune, Orlando Sentinel three US papers owned by Alden; they filed the second suit
   • Anthropic, Perplexity, Mistral, xAI and Cohere AI companies the suits name as buyers of Brave's index; they are not being sued here

   How it unfolded

   Brave was founded in 2015 and has raised about $250 million, by private-market counts. [1]

   By 2023 Brave had made little headway in search, so it began selling access to its index, the Times says. [3]

   In March 2025 Brave sued News Corp, asking a court to call its indexing 'fair use', copying US law allows. [2] News Corp sued back in July. [2]

   Last week OpenAI, Google, Microsoft and Anthropic backed a shared way for sites to refuse AI training. [5]

   1. Mar 2025 Brave asks a court to call its indexing fair use, in a fight with News Corp [2]
   2. Jul 2025 News Corp sues back, accusing Brave of theft [2]
   3. This week the papers sue Brave
   4. 2015 Brave is founded [1]
   5. 2023 Brave starts selling access to its index, the Times says [3]
   6. 2 Nov 2026 The Times and three Alden papers sue Brave in Manhattan [2]

   Why it matters

   The papers cite a study finding AI crawlers send back 95.7% less traffic than Google search. [2]

   Fewer visits mean fewer chances to sell adverts and subscriptions, which pay for the reporting. [2]

   The Times has also sued OpenAI and Perplexity over similar claims. [1]

   How it works

   Copyright law has a name for one of the papers' claims: contributory infringement, which means helping others copy a work. [2]

   Here the papers say Brave did the copying, then sold access on to 5 AI companies. [1] So this suit is aimed at the supplier, Brave. [1]

   Seen it before

   A shop that sells copied films can be in trouble too, not only the person who copied them.

   The truth

   • What is really going on The Times and three other papers sued Brave because, they say, it copied their pages and sold access to Anthropic, Perplexity, Mistral, xAI and Cohere. [2]
   • Every claim here is the papers' own, and Brave had not replied when the suits were reported. [1]
   • Nobody knows yet

     Whether a court will find Brave's copying legal.

     Brave asked a court in 2025 to call its indexing fair use, and the papers' new claims have not been tested. [2]

   • Nobody knows yet

     Which AI models learned from the papers' articles through Brave.

     The reporting ties training to one older model, Claude 2, through a Brave blog post. [2]

1. 02

   Anthropic says AI-sped progress may be near

   What happened

   • Anthropic, the company behind the Claude AI, updated its filing to sell shares on Monday. [18]
   • Its list of risks says it may soon meet its own test for 'dramatically accelerated progress attributable to AI'. [18]
   • That test is set out in its Responsible Scaling Policy, its own rules for handling risky AI. [18]
   • The filing says none of its current models has crossed that line yet. [18]

   How it unfolded

   It also says some past jumps in its models' abilities were spotted only after they happened. [18]

   Why it matters

   Anthropic says its models may improve faster than it can build the safety checks for them. [18]

   How it works

   AI researchers call this recursive self-improvement: AI that gets good at improving AI. [18]

   Anthropic already uses AI in its research. [18] So each better Claude can speed up work on the next one. [18]

   Seen it before

   A pupil who learns how to study better then learns every next subject faster.

   The truth

   • A share-sale filing both warns buyers and sells to them, so this claim may serve both purposes. [18]
   • Nobody knows yet

     When Anthropic will say a model has crossed its own line.

     Its filing rests on internal estimates about future models, and outsiders cannot check them. [18]

   What comes next

   It says it is considering rule changes to keep humans in control of building new models. [18]

2. 03

   Snap settles before its first addiction trial

   What happened

   • Snap, the company behind the Snapchat app, settled a lawsuit days before its trial in Los Angeles. [19]
   • The two who sued are S.J., 16, from Illinois, and K.D.B., 18, from Mississippi. [19]
   • They blamed features like autoplay and endless scrolling for depression, anxiety and eating disorders. [20]
   • Snap's boss, Evan Spiegel, will now not have to testify. [21]

   Who is involved

   • S.J., 16 Illinois
   • K.D.B., 18 Mississippi
   • Meta settled
   • TikTok settled
   • YouTube settled
   • Snap settled Monday

   How it unfolded

   They had already settled with Meta, TikTok and YouTube, so Snap was the last company left. [19]

   In March a Los Angeles jury found Meta and YouTube negligent in the first such trial. [19] It awarded $6 million, and both appealed. [19]

   Why it matters

   In August Meta agreed deals worth up to $18bn with nearly every US state. [19] Of that, $5bn depends on similar deals with Snap, TikTok and YouTube. [19]

   How it works

   Lawyers call this a bellwether trial: a few cases picked from thousands to test the claims first. [19]

   This would have been Snap's first trial in the wider fight. [19] Settling it means no jury tested the claims against Snap. [21]

   Seen it before

   A class tries the hardest question first, and how it goes tells everyone how the rest may go.

   The truth

   • Before settling, Snap said the claims misrepresent its app, which opens to a camera, not a feed. [22]
   • Nobody knows yet

     What Snap paid to settle.

     The terms are confidential. [19]

   What comes next

   The next test trials are due to start in March, the teenagers' lawyers said. [19]

1. 04

   Squarespace cuts about 300 jobs

   What happened

   • Squarespace, a company that helps people build and host websites, is cutting about 300 jobs. [23]
   • That is 19% of its 1,800 staff. [24]
   • Most cuts hit the teams that design and build its products, the boss's memo says. [6]
   • Anthony Casalena, the boss, wrote of 'smaller teams, more player coaches', meaning leaders who also do the work. [6]
   • Those leaving get at least 16 weeks of pay and health cover until May 2027. [24]

   How it unfolded

   Tools that build a website from a few typed words have upended website building, the work Squarespace sells. [23]

   Squarespace launched its own AI website builder, Blueprint, in 2024. [24] It went private that year in a $7 billion deal. [23]

   1. 2024 Blueprint AI tool launched
   2. 2024 $7 billion deal takes it private
   3. Now jobs cut

   Why it matters

   Last week Microsoft offered buyouts to 60,000 staff, and other firms also point to leaner teams. [6]

   How it works

   Why AI takes over single tasks first, and whole jobs later

   Labour economists see each job as a bundle of tasks: AI takes the routine ones first, and firms regroup the rest into fewer roles.

   The twist

   AI rarely deletes a whole job in one go. It takes the routine steps out of many jobs, and the firm then needs fewer people for the steps that are left.

   The picture

   9 of 10

   steps of AI agent work Gartner counts as routine

   Step by step

   1. 1 A job is a bundle of many tasks
   2. 2 AI is cheapest at the routine, repeated ones
   3. 3 Firms move those tasks to software
   4. 4 The tasks left over are regrouped into fewer roles
   5. 5 Some roles shrink while others change

   The same force, elsewhere today

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

   • SUSE cuts a tenth of its staff

     Its boss said some roles are needed far less and others will change, which is tasks moving rather than whole jobs vanishing.

   • Squarespace cuts about 300 jobs

     Most cuts hit the teams that build its products, and the boss wants smaller teams where leaders also do the work.

   • Gartner: cheap open models can do most agent work

     It split AI work into steps and found 9 in 10 routine enough for a cheap model, leaving 1 in 10 for the costly one.

   • Heavy AI users work longer days

     Time saved on one task was filled with more tasks, so the day grew instead of shrinking.

   Where you've seen this

   Bank cash machines

   machines took over counting out cash, and bank staff spent more time on loans and advice

   Self-checkout tills

   scanning moved to the shopper, and fewer staff now watch several tills at once

   Spreadsheets

   adding up columns went to the computer, and bookkeepers moved to checking and planning

   The catch

   The task model says which work a machine can do. It cannot tell you whether a firm cut jobs because of AI or because its owners wanted lower costs.

   And the whole of it

   The engineers at SUSE and Squarespace, the managers reading Claude's reports and the staff whose hours ActivTrak counts each see only their own tasks change. Most people who work at a desk are inside this too, with their own bundle of tasks being sorted one step at a time.

   The truth

   • What is really going on On the same day SUSE and Squarespace said about 490 jobs would go, and both bosses pointed to AI or to smaller teams. [7][6]
   • Who pays Squarespace's product and engineering teams — Most of the 300 cuts fall on them, with 16 weeks of pay at least. [6]

1. 05

   SUSE cuts a tenth of its staff, citing AI

   What happened

   • SUSE, a German company that makes Linux software for businesses, is cutting about 10% of its staff. [7]
   • It has more than 1,900 employees, so about 190 people will go. [7]
   • Its boss, Dirk-Peter van Leeuwen, told staff AI and automation are changing how software firms work. [25]
   • 'The requirement for some existing roles has greatly diminished,' he wrote. [25]

   How it unfolded

   Swedish investment firm EQT bought SUSE in 2018 and took it off the stock market in 2023. [7]

   1. 4 Nov affected staff told
   2. 16 Nov new plan shown to all

   Why it matters

   Van Leeuwen called the cuts painful for SUSE's people, its culture and its open source community. [7]

   How it works

   In organisation design, a flatter firm has fewer layers of managers, so each manager leads more people. [25]

   Van Leeuwen says new tools let SUSE's teams work 'with less operational overhead and flatter reporting lines'. [25]

   Seen it before

   A school with fewer heads of year has each one looking after more classes.

   Its boss said some roles are needed far less and others will change, which is tasks moving rather than whole jobs vanishing.

   The idea, in story 04: Why AI takes over single tasks first, and whole jobs later →

   The truth

   • SUSE has posted no public statement; the news comes from a memo The Register saw. [7]

   What comes next

   Staff affected will be told by 4 November, and everyone sees the new plan on 16 November. [25]

2. 06

   Putin orders data centres to keep room for the state

   What happened

   • Russia's president, Vladimir Putin, ordered private data centres to hold back 10% of their space for the state. [26]
   • If called on, owners must hand that space over within 24 hours. [26]
   • A new government commission will decide when to call on it. [26]

   How it unfolded

   Ukrainian drones have hit Russian data centres in recent weeks. [26]

   In early October they struck Russia's largest, a 63-megawatt site, and users of Yandex mail and payments saw problems. [26]

   Later Russia's finance ministry's data centre, said to hold tax records, was damaged too. [26]

   Why it matters

   The aim is to give the state more computer space quickly, if it needs it. [26]

   How it works

   Engineers call this spare capacity: room kept free so work can move when one site fails. [26]

   Drones damaged the data centre said to hold Russia's tax records. [26] Space held free elsewhere gives such work somewhere to move. [26]

   Seen it before

   A spare key at a neighbour's means one lost key does not lock you out.

3. 07

   A prosecutor cites a case that does not exist

   What happened

   • A senior prosecutor cited a court case that does not exist, in England's High Court. [27]
   • She works for the Crown Prosecution Service, which brings criminal cases in England and Wales. [27]
   • The fake case came from an AI summary in an internet search, the Guardian understands. [27]
   • It was cited to argue that a child should be prosecuted. [27] Judges said the error did not change the outcome. [27]

   How it unfolded

   It is the second known case like this at the CPS; the first came out in February. [27]

   A tracker kept by barrister Matthew Lee now lists 101 confirmed, suspected or disputed AI errors in UK courts. [27]

   Why it matters

   'That's a serious failure,' said Brian Flanagan, a law professor who studies judges' use of AI. [27]

   How it works

   AI researchers call this a hallucination: a chatbot writes something that sounds right but is invented. [27]

   The invented case was given a real-looking name, R v J, and a 2024 date. [27] So it passed as real until someone checked. [27]

   Seen it before

   A friend confidently gives you directions to a shop that closed years ago.

   The truth

   • The CPS says its rules demand human checks, and that the error came from internet research. [27]
   • Nobody knows yet

     How many other court papers rest on AI-made cases.

     Matthew Lee says his tracker of 101 may be the tip of the iceberg, because many lower court decisions are never published. [27]

4. 08

   Nine in ten US data-centre comments are against

   What happened

   • A new database of 10,202 comments at US council meetings this year found 90.5% against data centres. [15]
   • Only 7% spoke in favour, and 73% of meetings heard no supporter at all. [15]
   • The site, American Sentiment, is run by Prophecy Gov, a San Francisco AI start-up. [15]
   • Water use was the top worry, and only 21% of comments mentioned AI. [15]
   • Red and blue counties objected at almost the same rate. [15] Of commenters who said they were union workers, 79% were in favour. [15]

   Why it matters

   In Texas, the Democratic candidate for governor promises a strict pause on new data centres. [28]

   A pause there could affect about 29% of the new US data-centre capacity planned for 2028. [28]

   How it works

   Planners call this 'not in my back yard': people may accept something in general but fight it near home. [15]

   Neighbours carry the water use and the change to local life. [15] Many told meetings they doubted the promised benefits would come. [15]

   Seen it before

   Everyone wants a bus stop on the street, but not right outside their own window.

   The truth

   • Comments come from people who turn up, and the site says they may not match the whole country. [15]
5. 09

   Google's Minnesota data centre wins a vote

   What happened

   • Hermantown, a small city in Minnesota, approved a Google data centre on 30 October. [16]
   • The council vote was unanimous, after at least 2 delayed meetings and hours of objections. [16]
   • To win it, Google cut the plan from 4 buildings to 3 and capped their height at 33 feet. [16]
   • It also dropped a request for up to $80 million in property tax breaks. [16]
   • The city now expects about $3 million a year in property tax. [16]

   Why it matters

   The site needs about 500 megawatts. [16] A consumer group says that is close to all the homes in Minneapolis, St Paul, Duluth and Rochester. [16]

   How it works

   In land-use planning, a special use permit lets a council attach conditions before it says yes. [16]

   Here the conditions include 3 buildings, a height cap and checks on nearby wells. [16] A $210 million deal covers clearing the site if it closes. [16]

   Seen it before

   Parents allow a party if you agree to finish early and clean up after.

   The truth

   • Council member John Geisler said the vote was 'not about big tech' but about land use. [16]

   Good news

   • Hermantown in Minnesota expects about $3 million a year in tax from Google's new data centre, after Google dropped its request for up to $80 million in tax breaks.

   What comes next

   Three lawsuits still challenge the project, and Minnesota's utility regulator must still approve its power. [16]

6. 10

   OpenAI says its power plants can help the grid

   What happened

   • OpenAI says power plants at its planned data centre in Effingham County, Georgia, will help the grid in emergencies. [17]
   • It promises to free up to 1,000 megawatts in extreme weather, enough to cool about 700,000 homes. [17]
   • It calls this its 'Grid Protectors Program', and now calls its data centres 'compute campuses'. [17]

   How it unfolded

   Tennessee, Florida, Oklahoma and Oregon have laws to stop AI firms passing energy costs on to residents. [17]

   Why it matters

   OpenAI argues that steady year-round demand spreads the grid's fixed costs over more power sold. [17]

   How it works

   Grid engineers call this demand response: a big user agrees to cut its use when the grid is strained. [17]

   OpenAI's site would draw power all year, but cut back on the days when homes need it most. [17]

   Seen it before

   A family skips the oven when everyone in the street is cooking at once.

   The truth

   • The pilot is promised for a data centre still to come, so none of this power has been freed yet. [17]
7. 11

   A 16GB Raspberry Pi now costs $385

   What happened

   • Raspberry Pi raised the price of its 16GB Raspberry Pi 5, a small cheap computer, by $80 to $385. [12]
   • That is more than triple the $120 it cost at launch. [12]
   • It is the 5th price rise since October 2025. [29]
   • Eben Upton, the boss, blamed a lack of other suppliers for 16GB memory chips. [29]

   How it unfolded

   Since late 2025, AI data centres buying memory have pushed up prices for everyone else. [30]

   A 32GB memory kit cost $94.99 in 2020. [30] The cheapest similar kit cost $198 this October. [30]

   Why it matters

   Makers are moving to cheaper boards for small projects and mini PCs for big ones. [12]

   How it works

   Economists call this a demand shock: a big new buyer arrives faster than supply can grow. [30]

   AI data centres now bid for the same memory chips. [30] So a small maker like Raspberry Pi pays the higher price. [29]

   Seen it before

   When a whole team queues for the last few snacks, the tuck shop can charge more for each one.

   The truth

   • Who pays Hobbyists and makers buying a big Raspberry Pi — Each memory price rise reaches them straight away, with no other 16GB chip to switch to. [29]
8. 12

   Windows 11 upgrades stall as PCs get dearer

   What happened

   • People and firms have slowed their move to Windows 11 as chips and memory get dearer, analysts say. [13]
   • Microsoft's Windows sales to PC makers fell 27% in its latest quarter. [13]
   • Windows 10 still ran on 27.7% of Windows PCs in September, up from 26.45% in February. [13]

   How it unfolded

   Microsoft stopped mainstream support for Windows 10 a year ago. [13]

   Intel has also raised its processor prices, adding to the cost of each new PC. [13]

   1. 2028 relief may start
   2. 2029 or even later

   Why it matters

   PC makers are still sitting on unsold stock, Microsoft's finance chief said. [13]

   How it works

   Economists call this a deferred purchase: when prices rise, people keep the old thing longer. [13]

   'When prices are higher, demand goes down,' said analyst Dean McCarron. [13] So more than a quarter of Windows PCs still run Windows 10. [13]

   Seen it before

   When school shoes get dearer, families make last year's pair last another term.

   What comes next

   Relief may not come until 2028 or 2029, analysts said. [13]

9. 13

   Musk cuts Grok Bot use over costs

   What happened

   • Elon Musk's SpaceX cut how much people can use Grok Bot, its AI helper, without warning. [9]
   • Some users' weekly allowance jumped from 50% used to 100% overnight. [9]
   • 'The cost of serving Grok Bot was unsustainably high relative to revenue,' Musk wrote on X. [9]

   How it unfolded

   SpaceX, which now owns Musk's AI business, says AI makes up 52% of its income. [9]

   Its AI division made a $431 million profit, which is rare in the AI industry. [9]

   Why it matters

   Musk has called Grok Bot the fastest-growing product he has ever seen. [9]

   How it works

   Economists call this marginal cost: what it costs to serve one more request. [9]

   Each Grok Bot task costs SpaceX computer time. [9] So a fixed monthly fee loses money on the heaviest users. [9]

   Seen it before

   An all-you-can-eat buffet loses money on the diner who goes back again and again.

   The truth

   • Who gains SpaceX's AI business — Tighter Grok Bot limits cut its serving bill without a higher fee for subscribers. [9]
   • Who pays Grok Bot subscribers — They pay the same fee for less use, cut overnight. [9]
   • Nobody knows yet

     How Grok Bot's limits will change from here.

     Musk has not said; his only word came in a reply to one user on X. [9]

10. 14

    Gartner: cheap open models can do most agent work

    What happened

    • Gartner, a research firm that advises companies, says 9 in 10 steps of AI agent work are routine. [11]
    • Those steps include sorting items, calling other software and checking data. [11]
    • Open-weight models, which anyone can download and run, do those steps as well as paid ones, it says. [11]
    • They cost a quarter of the cheapest paid model. [11] Against a top model, they can cost a twentieth. [11]
    • In practice the saving is closer to 5 or 6 times, because open models retry more. [11]

    Why it matters

    That would leave the big AI labs earning from only about 10% of the steps. [11]

    Gartner adds that every step sent to an outside AI moves some of a firm's own records outside too. [11]

    How it works

    It split AI work into steps and found 9 in 10 routine enough for a cheap model, leaving 1 in 10 for the costly one.

    The idea, in story 04: Why AI takes over single tasks first, and whole jobs later →

    The truth

    • Who gains Makers of open-weight AI models — Gartner tells companies to send routine agent steps to models they can run themselves. [11]

    What comes next

    Gartner expects 90% of these steps to run on open models by 2030. [11]

11. 15

    Google lets a union into its UK offices

    What happened

    • Google agreed to let the Prospect union into its UK offices to sign up staff. [31]
    • It covers about 7,000 workers, including YouTube but not the AI lab DeepMind. [32]

    Who is involved

    • Meta deal in September
    • Google deal now
    • CWU union
    • Unite union

    How it unfolded

    A new UK law, in force since Friday, gives unions a legal right to enter workplaces. [31]

    Employers that keep refusing can be fined up to 500,000 pounds. [31]

    Prospect signed a similar deal with Meta in September. [31]

    Why it matters

    If enough staff join, Prospect can ask for a say on pay, flexible work and how Google uses AI. [31]

    How it works

    Employment law calls it union recognition: once recognised, a union bargains with the employer for staff. [31]

    Access comes first: Prospect can now talk to staff in person and online, to sign up enough members. [31]

    Seen it before

    A school club can only start once enough pupils sign the list.

    The truth

    • Who gains The Prospect union — With Meta and now Google signed, its leader says it can show other tech firms it wins deals with even the biggest. [32]

    What comes next

    Two other unions, the CWU and Unite, are seeking recognition at DeepMind. [32]

12. 16

    OpenAI signs deals in Abu Dhabi

    What happened

    • OpenAI's boss, Sam Altman, signed 4 deals in Abu Dhabi, the capital of the United Arab Emirates. [33]
    • The biggest is with ADNOC, the state oil company, to run AI agents across its oilfields, refining and trading. [33]
    • The Abu Dhabi government will make some of its policies readable by AI agents. [33]
    • Mubadala, a state investment fund, will use OpenAI's models to study possible deals. [33]

    How it unfolded

    OpenAI is raising $20 billion, and investors from the UAE are co-leading the round. [34]

    Why it matters

    Asked on stage about his biggest challenge, Altman gave one word: 'Energy.' [33]

    The truth

    • No money figures were given for the ADNOC deal. [33]
13. 17

    OpenAI hires a former US Army chief

    What happened

    • OpenAI hired Dan Driscoll, a former US Army Secretary, the civilian who runs the army, in a new security job. [35]
    • He will advise on using AI for US security and cybersecurity, and help allies adopt it. [35]
    • Altman spoke of a chance to spread 'democratic AI' with other countries. [35]
    • Driscoll said AI could cut paperwork and get answers faster to people in national security. [35]

    Why it matters

    Security needs 'are changing faster than at any time in the past fifty years,' Driscoll said. [35]

    How it works

    Political scientists call it the revolving door: people move from government into firms that sell to government. [35]

    Driscoll ran the US Army. [35] Now he will help OpenAI work with the US and its allies on security. [35]

    Seen it before

    A referee who becomes a coach takes their inside view of the rules to the new side.

14. 18

    Phone masts learn to spot drones

    What happened

    • Ericsson, a Swedish maker of mobile-network gear, says phone masts can spot and track drones. [36]
    • A signal from one mast bounces off a drone and is caught by several others. [36]
    • Small trials start in 2027, with wider use from 2028. [36]
    • AT&T, a US phone company, tested it in Texas, and so did Finland's emergency network. [36]
    • NATO tested tracking drones with existing 5G masts in Latvia from 19 to 30 October. [37]

    Who is involved

    • AT&T Texas
    • Finland's emergency network Finland

    How it unfolded

    1. 2027 small trials
    2. 2028 wider use
    3. 19 Oct NATO test starts in Latvia
    4. 30 Oct test ends

    Why it matters

    Cheap drones keep closing airports; Jersey's shut over the Halloween weekend. [36]

    How it works

    Engineers call it integrated sensing and communication: one radio both carries calls and senses objects. [36]

    Masts already cover the low air where drones fly, which ordinary radar sees poorly. [36]

    Seen it before

    A bat finds a moth by the echo of its own squeak.

    The truth

    • Tom's Hardware notes that masts could become bigger targets in a war. [37]
    • Who gains Ericsson — Drone sensing lets it sell a new job to phone networks it already supplies. [36]

    Good news

    • In tests in Texas and Finland, ordinary phone masts found several drones at once and followed where they flew.
15. 19

    Ukraine watches Russian drones' own video

    What happened

    • CODE380, a start-up in Kyiv, makes a pocket receiver that shows live video from many Russian drones. [38]
    • A Ukrainian unit can see where a drone is heading and what its pilot is aiming at. [38]
    • The device weighs about 10 ounces, less than half some rival Ukrainian models. [38]

    How it unfolded

    It also listens on channels Russia's newer Molniya drones began using in June, which most detectors miss. [38]

    Why it matters

    Serhii Beskrestnov, a radio expert who advises Ukraine's president, warned that trusting one detector too much could be fatal. [38]

    How it works

    Engineers say a signal is sent in the clear when nothing scrambles it. [38]

    These drones send their video unscrambled. [38] So anyone with the right receiver can watch along. [38]

    Seen it before

    On a walkie-talkie, anyone on the same channel hears what you say.

    The truth

    • If the drones switch to digital video, receivers like this one would go blind, Beskrestnov has warned. [38]
16. 20

    Russia hides jammers against Starlink

    What happened

    • Russia is putting jammers on trailers to cut Ukraine's drones off from SpaceX's satellites, Beskrestnov says. [39]
    • Each jammer covers about 25 square kilometres, he said. [39]
    • He said Russia has covered nearly all important sites in Crimea this way. [39]
    • He told drone pilots to watch truck and trailer parks, where jammers can hide. [39]

    Who is involved

    • Starshield military Starlink
    • Ukraine's drones steered through it

    How it unfolded

    Ukraine's forces use Starshield, the military version of SpaceX's Starlink, to steer strike and scout drones. [39]

    Why it matters

    Starshield has been crucial to Ukraine's control of mid-range attack drones, Business Insider reports. [39]

    How it works

    Militaries call this electronic warfare: attacking the radio links that weapons depend on. [39]

    A jammer drowns out the satellite link with noise. [39] So pilots far away can no longer steer their drones. [39]

    Seen it before

    In a noisy hall, the louder the crowd, the less your friend hears you.

17. 21

    Claude reports show bosses what AI costs

    What happened

    • Anthropic launched 'smart reports' that show companies what staff use Claude for and what it costs. [8]
    • Claude reads a sample of a team's sessions, groups the work, and puts a cost on each group. [8]
    • The reports can flag sessions with 'no work value' or that were 'personal or off-topic'. [8]
    • Each company can run up to 10 reports a month for free. [8]
    • Names are hidden by default, but an owner can switch them on, and each reveal is logged. [8]

    Why it matters

    Goldman Sachs limited Claude Code to its engineers after weighing productivity and risk, Bloomberg reported. [8]

    The truth

    • Anthropic says the reports must not be used to judge staff or make job decisions. [8] But it lets owners show names. [8]
18. 22

    Heavy AI users work longer days

    What happened

    • ActivTrak, which sells software that tracks staff activity, found that heavy AI users work longer days. [40]
    • They worked 8.42 to 9.08 hours a day, against 7.58 for people not using AI. [40]
    • Their focus sessions were 22% shorter, and they switched apps 15% more often. [40]

    How it unfolded

    Its earlier report found the opposite: AI users' workdays had got about 9 minutes shorter. [40]

    The new study covers 9,140 workers over 5 months; the old one covered 163,638 over 3 years. [40]

    Why it matters

    When staff finish tasks faster, managers may hand them more work, ActivTrak's researcher said. [40]

    How it works

    Time saved on one task was filled with more tasks, so the day grew instead of shrinking.

    The idea, in story 04: Why AI takes over single tasks first, and whole jobs later →

    The truth

    • An analyst pointed out that ActivTrak sells the tool that makes these measurements. [40]
    • Nobody knows yet

      Whether AI lengthens or shortens the working day.

      ActivTrak's two studies point opposite ways, and they used very different samples. [40]

19. 23

    Y Combinator's boss spends up to $200,000 a year on AI

    What happened

    • Garry Tan, who runs the start-up investor Y Combinator, says he spends $100,000 to $200,000 a year on AI. [10]
    • He pays for tokens, the pieces of text AI companies charge by. [10]
    • His AI agents have read his email, his Dropbox files and 20 years of therapy notes, he said. [10]
    • He shares his set of 23 Claude Code skills, called gstack, free on GitHub. [10]
    • Y Combinator also uses an agent that is 'rebuilding the way we operate at YC', he said. [10]

    Why it matters

    He said people too worried about safety lose the best parts of working with an agent. [10]

    The truth

    • The spending figure is Tan's own rough guess, 'maybe' $100,000 to $200,000, given on a podcast. [10]
20. 24

    China's robot makers race for AI brains

    What happened

    • Chinese robot makers that mastered running and backflips are now racing to build the AI that controls them. [41]
    • Chinese robot-AI companies raised 93.5 billion yuan, about $13 billion, in the first half of 2026. [41]
    • That was 5 times as much as a year earlier, across 322 deals. [41]
    • UBTech, a robot maker, advertised a chief scientist job paying up to 124 million yuan, about $18 million. [41]
    • CATL, a battery maker, says Spirit AI robots plug in wires at its plant with over 99% success. [41]

    Why it matters

    US export rules limit China's access to Nvidia's best chips, which robot AI needs. [41]

    How it works

    Economists call it commoditisation: when many rivals make the same thing, its price falls. [41]

    Robot bodies are now cheap, so one consultant says the profit will come 'from the brain'. [41]

    Seen it before

    Once every shop sells the same phone case, the shop with the best design wins.

    Good news

    • Robots from Spirit AI now plug in wires at a CATL battery plant in China, and get it right more than 99% of the time.
21. 25

    AI firms set up election war rooms

    What happened

    • OpenAI and Anthropic each set up an election 'war room' for the US midterms, a first for both. [42]
    • The teams watch their chatbots for wrong election information and misuse. [42]

    Who is involved

    • OpenAI war room
    • Anthropic war room

    How it unfolded

    Meta set up the first tech war room in 2018, over false election posts. [42]

    Last week Google's AI Mode would not say who won the 2020 US election; Google said it was fixing that. [42]

    1. 2018 Meta's first war room
    2. Now OpenAI and Anthropic join in

    Why it matters

    AI search tools tend to reflect back the results users want to see, which makes the job harder. [42]

    The truth

    • A former Meta official said war rooms exist for the things 'we can't necessarily predict going wrong'. [42]
22. 26

    Meta ends Quest 2 updates

    What happened

    • Meta will send the last software update for its Quest 2 and Quest Pro headsets by March 2027. [43]
    • Security fixes continue until the end of 2027, and app makers can submit apps until December 2027. [43]

    Who is involved

    'Millions of people still play on them,' Meta says. [43]

    How it unfolded

    Meta stopped selling both virtual-reality headsets in 2024. [43]

    1. Mar 2027 last software update
    2. Dec 2027 last security fixes and new apps

    Why it matters

    Meta wants app makers to build for the Quest 3 and its new VR Glasses, due in spring 2027. [43]

    How it works

    Engineers call this end of support: the maker stops changing the software, then stops fixing it. [43]

    After 2027, Quest 2 owners get no more security fixes, even for holes found later. [43]

    Seen it before

    An old textbook still works, but any mistakes in it stay there for good.

    The truth

    • Who pays Quest 2 and Quest Pro owners — To keep getting security fixes after 2027, they will need a newer headset such as the Quest 3. [43]
23. 27

    Patients can show Claude their health records

    What happened

    • Athenahealth, whose software serves over 20% of the US population, now links patients' records to Claude. [44]
    • Patients sign in, choose what Claude can see, and can take access back. [44]
    • It works only for paying Claude Pro and Max users, at $20 and $100 a month. [44]
    • It uses MCP, a standard way for an AI to plug into other software. [44]

    Why it matters

    Patients can then ask Claude questions about their own records. [44]

    The truth

    • That the link is secure is Athenahealth's own claim. [44]
24. 28

    ChatGPT merges its chat and work tabs

    What happened

    • OpenAI will merge the Chat and Work tabs in ChatGPT's desktop app, starting in December. [45]
    • Plus, Pro and Business plans will share one usage allowance across ChatGPT and Codex, its coding tool. [45]
    • Users will be able to ask ChatGPT to build a website or a presentation, and review its work. [45]
    • Codex will stay a separate part of the app. [45]

    Why it matters

    OpenAI says users can go 'from exploring an idea to having ChatGPT carry it through'. [45]

    The truth

    • With one shared allowance, time spent coding in Codex now counts against the same limit as chatting. [45]

    What comes next

    Some users will try the new layout first, before it reaches everyone. [45]

25. 29

    Claude agents stop resending whole chats

    What happened

    • Anthropic launched message threads for developers building AI agents on Claude. [46]
    • An agent sends the whole conversation once, then only what is new. [46]
    • On very long sessions, the wait before Claude replies fell about 35%, Anthropic says. [46]

    How it unfolded

    Anthropic cut its cache prices by 75% in September. [46]

    Why it matters

    A coding agent can make hundreds of requests in one session, each carrying every earlier message. [46]

    The truth

    • Prices stay the same, because Claude still reads the whole conversation every turn. [46]
26. 30

    Canada's clock change trips up Microsoft

    What happened

    • Parts of Canada stopped changing their clocks on 1 November, and Microsoft's software has not caught up. [47]
    • Areas of British Columbia, Alberta and Manitoba now stay on one time all year. [47]
    • Outlook meetings set after 1 November may be an hour out until users fix them by hand. [47]
    • One workaround tells some software to act as if it were in Saskatchewan, Arizona or Bogota. [47]
    • One admin on Reddit said they wrote about 900 lines of code to fix colleagues' calendars. [47]

    How it unfolded

    1. 10 Nov SharePoint fix
    2. 12 Dec Power Pages fix

    Why it matters

    Recurring meetings are worst: changing the time zone resets moved ones, which must be made again. [47]

    The truth

    • Microsoft published its guidance on 26 October, less than a week before the change. [47]

    What comes next

    A fix for SharePoint is due on 10 November, and one for Power Pages on 12 December. [47]

27. 31

    A free stand-in for a VMware tool

    What happened

    • A cloud architect, Andrija Panic, built a free replacement for a VMware tool that VMware pulled from public download. [48]
    • VMware makes software that lets one computer act as many, called virtual machines. [48]
    • The tool, VDDK, reads virtual machines' disks, which other tools need in order to move them. [48]

    Who is involved

    Tools from rivals such as Proxmox and Apache CloudStack rely on that same small library. [48]

    How it unfolded

    In August VMware limited downloads to its partners, so its customers could no longer get it. [48]

    1. August downloads limited to partners
    2. Now a free replacement

    Why it matters

    'People's ability to move or protect their own data' should not depend on a download page, Panic said. [48]

    How it works

    Economists call this a switching cost: anything that makes leaving a supplier harder. [48]

    Pulling the download raised that cost for every customer trying to move off VMware. [48]

    Seen it before

    A phone that charges only with its own maker's cable is harder to swap.

    Good news

    • His replacement is open source, so anyone can use it to move their machines. [48]
28. 32

    Chip start-up Tachyum files for bankruptcy

    What happened

    • Tachyum, a chip start-up, filed for bankruptcy in Bratislava, Slovakia, on 22 October, Forbes reported. [49]
    • It never showed a working Prodigy chip after 8 years of work. [49]

    How it unfolded

    It announced Prodigy in 2018 and promised it for 2020, then pushed it back year after year. [49]

    Its last redesign, in late 2025, pushed sales back to 2030 or later. [49]

    1. 2018 Prodigy announced
    2. 2020 first promised date
    3. 2030+ date after the last redesign

    Why it matters

    Its designs and patents could still be bought, so Prodigy might live on under a new owner. [49]

    How it works

    Project managers call it scope creep: the target keeps growing, so the project never ships. [49]

    Prodigy grew from 64 cores to 128, then 192, then 1,024, each needing new work. [49]

    Seen it before

    A school project that keeps getting a new chapter added is never handed in.

    What comes next

    If a court declares it bankrupt, a trustee could sell its designs and patents to repay lenders. [49]

29. 33

    A Toyota-backed chip keeps its AI on board

    What happened

    • AISIN, a car-parts maker whose biggest owner is Toyota, showed a chip that keeps its AI model on board. [14]
    • Most chips fetch AI models from separate memory, and the back-and-forth makes tasks like spotting objects slower. [14]
    • The chip starts up in 65.7 milliseconds, 38 times faster than with outside memory. [14]
    • Squeezing the model shrank it by 53%, from 17.7MB to 9.36MB. [14]

    Who is involved

    It was built with researchers at Tohoku University in Japan. [14]

    Why it matters

    Outside memory chips are now very expensive, which makes building memory in more attractive. [14]

    The truth

    • AISIN has not said if or when the chip will be made. [14]

    Good news

    • A new chip from AISIN and Tohoku University starts up in 65.7 milliseconds, 38 times faster than chips that keep their AI in outside memory.