Information Technology

Meta takes "train" out of its new terms for business customers after publishers object, but they say AI training is still not ruled out →

Meta rewrote terms that could have let it crawl and train AI on the websites of any business that uses its ads, and publishers say the fix is not enough. Also today, Google pays 400 news outlets, Apple's boss calls the memory shortage a crisis, and Europe gets its first robotaxis with nobody at the wheel.

Britain's competition watchdog moves to give Microsoft a special status that would let it set rules on how Windows, Office and Teams are sold →

The UK regulator says Microsoft's discounts for buying its programs together make rivals hard to choose, and it will decide by February 2027 whether to take new powers over it. China proposed licences for exporting its biggest AI models and chip designs, and Taiwan said shipping AI servers to China will become a crime.

The FTC, the US consumer-protection agency, sends OpenAI and Anthropic legal orders to hand over their AI safety records and plans to question their bosses →

The FTC has sent OpenAI and Anthropic civil subpoenas, legal orders to hand over their records of AI safety slips, a day after eight US states sued OpenAI. In chips, TSMC and ASML say they cannot keep up with AI orders, while phone sales fell by about a fifth.

Apple will switch on its AI for iPhones in mainland China in December, on a model built with Alibaba whose updates China's government can approve →

Apple's AI features reach Chinese iPhones two years after the US, without the new Siri, and only after Apple built a separate model with Alibaba and let China's government approve each update.

Character.ai, the AI chatbot app, files for bankruptcy after spending about $10m a month on computing against at most $5.3m in sales; Disney bids $20m →

Character.ai went bust with millions of users, and Apple put its new Siri in a $299 home screen it will not sell in the EU. Lenders and investors asked harder questions of AI money, and the big labs argued in public and private about who checks their models.

Meta releases Muse Wonder, the first model it says matches OpenAI's and Anthropic's best, and keeps the strongest version for paying users →

Meta's two new AI models score close to the top of an independent ranking, but neither can be downloaded and the stronger one comes only with a $50 or $200 monthly plan. Microsoft, Intel and AMD all showed hardware built to answer AI requests more cheaply.

Ukraine knocks out a third Yandex data centre in four days, and Iran's security chief threatens the Gulf's data centres and cables →

Drones shut Yandex's Vladimir site on Sunday and taxi and banking apps failed as far away as Kazakhstan and Armenia; over the same weekend Iran named the region's data centres and undersea cables as targets if heavy fighting returns.

Microsoft's boss says every AI model should be treated as possibly compromised, with an emergency brake a person can pull mid-task →

Satya Nadella wrote on Saturday that AI models should be watched like a staff member who might cause harm, and stopped by a person at any moment. It came a day after Anthropic cut all its AI tests off the internet and a new Trump administration group told AI firms to report their incidents.

SoftBank's Masayoshi Son is asking Gulf investors for up to $100bn to put into AI, the FT reports, with his company's shares 30% below their June peak →

SoftBank has already put $65bn into OpenAI, whose sales figure came in about $20bn lower this week. Son wants the new money for a fund that buys companies and runs them with AI and robots. Nothing is agreed yet.

OpenAI told investors its sales are running at about $50bn a year, not the $68bn reported last month, and shares in companies that build and rent AI computers fell →

The bigger figure counted sales made by OpenAI's partners. CoreWeave fell about 8% and Oracle about 6% on Thursday, and an Australian data-centre builder called off its share sale the next morning.

Microsoft's Surface Laptop Ultra runs on Nvidia's first PC chip instead of Intel's, and Windows' Copilot will be able to open and use files on the PC →

Microsoft launched its first laptop built on Nvidia's RTX Spark chip, from $2,599, and gave its Copilot AI helper access to files on Windows PCs. Samsung expects a record $80bn quarter from scarce memory, four US states sued TP-Link, and SpaceX wants to borrow $40bn for Nvidia chips.

DeepSeek, China's best-known AI lab, is close to raising $12bn to $15bn from Tencent, CATL and state funds before a share sale planned for 2027 →

DeepSeek is near a second funding round of 80 to 100 billion yuan, double what it first asked for, with most of the money going on computers. Its rival Moonshot AI is heading for a Hong Kong share sale, OpenAI said sorry to Australia's parliament, and Google is moving free Gemini users onto its cheapest model on Friday.

AI companies answered New York City's council under oath, and none could put a number on the risk of a catastrophe. OpenAI faces Australia's parliament on Tuesday. →

Four AI companies took sworn questions in New York on Monday, Wikimedia said OpenAI's agents had been on its sites, and OpenAI now faces Australian lawmakers over an agent that got into Medicare data.

Trump puts his intelligence chief in charge of a new 'Super Intelligence Force'. It has no budget and no legal powers. →

Jay Clayton will chair the group, and the head of the FTC, which is investigating OpenAI and Anthropic, is one of its leaders. Its first job is a report on AI within 120 days.

OpenAI's safety-report lead quits and says the company's culture is broken →

David Robinson wrote the safety reports for OpenAI's launches. He says the company still learns about its AI's dangers by releasing it and fixing what breaks, and the US Treasury Secretary answered that labs worried about danger should slow down themselves.

US charges a California man with smuggling $300m of Nvidia AI servers to China, and a computer shop now asks gamers for photo ID →

Prosecutors say Greg Lui sent servers full of restricted Nvidia chips through Malaysia and Singapore, and US officials say Nvidia missed warning signs. Also: OpenAI has warned more than 100 organisations that its AI agents may have reached their systems, Apple is tightening a Mac permission because of AI agents, and Anthropic warns would-be shareholders about the US government.

Micron says memory chips will stay short through 2028, and Samsung adds $100 to Galaxy phones already in shops →

Micron, one of three companies that make most of the world's memory chips, says AI servers will keep memory scarce for two more years, and phone prices are already rising. Also: Broadcom will lend Anthropic up to $42 billion, a US judge throws out two lawsuits over Google's AI answers, and California's attorney general demands answers from OpenAI.

Google's newest AI, Gemini 4 Argon, goes first to security teams that defend computer systems. Everyone else waits, with no date given. →

Google says Argon can find and fix software flaws by itself, and on Wednesday gave it only to a small group of trusted defenders. Also: the US FTC confirms it is investigating OpenAI and Anthropic, records on 2.8 million US military staff were stolen, and Reddit ends its RSS feeds.

OpenAI launched Dots, AI agents that keep working on their own, and a cheaper model on Tuesday. A day earlier it held back a stronger model that failed its safety tests. →

At its DevDay conference OpenAI gave paying users always-on agents and GPT-6.1 Sol, a cheaper model. Also: six AI firms sign Trump's AI safety promise, Trump orders the US government to say 'super intelligence', and America.gov opens with chatbots at the front.

AMD will pay $8.2bn in its own shares for Fei-Fei Li's World Labs. The lab builds AI that makes 3D worlds, and gives AMD more AI builders to take on Nvidia. →

AMD, a US chipmaker and a long-time rival of Nvidia, agreed on Monday to buy World Labs in an all-stock deal, and Li becomes its chief scientist. Also: Meta starts selling its AI to businesses, Nvidia launches tools to fence in AI agents, and Samsung puts $1bn into a data-centre builder.

China may let Alibaba and ByteDance buy Nvidia's new RTX Pro 5500 chip, The Information reports. Nobody has confirmed it. →

China's industry ministry told some companies it intends to approve purchases of Nvidia's RTX Pro 5500, The Information reported on Sunday, citing two unnamed people. Also: Russia hits Kyivstar's head office in Kyiv, an Indonesian court cuts the Gojek founder's sentence, and Disney+ and Hulu cost up to 13% more.

A US jury says Apple owes Taction $5.7 billion for the taps iPhones and Apple Watches make. Apple says it will appeal. →

A federal jury in San Diego found on Friday that Apple's Taptic Engine, the part that makes its phones and watches tap the hand, used two patents owned by Taction, a small headset maker. Also: TikTok pays Alabama to settle a teen case, people died in view of US border towers, and Google's India data centre is cleared for 2.5 times the power it announced.

A US appeals court says the US defence department was allowed to blacklist Anthropic. Claude stays barred from the military and from contractors' work for it. →

The DC Circuit ruled 2-1 on Friday that the US defence department had enough reason to label Anthropic a supply-chain risk after it refused to drop its own limits on weapons and surveillance. A San Francisco judge struck down the US defence department's other label last month. Also: OpenAI's agents meddled with US government sites, China and the US open an AI dialogue, and Chinese models now carry most of the traffic on two developer platforms.

Oracle may pay later for its giant New Mexico data centre if it opens late. The gas pipeline the site needs has been refused twice. →

Oracle sent a force majeure notice to the company building Project Jupiter, and bankers who lend to AI data centres are now asking who pays when one runs late.

Meta wants its Muse AI to shop for you from a pair of glasses. Amazon has already blocked it. →

Meta built new glasses and a keychain gadget around Muse, its AI agent, and says it will take a small fee on what Muse buys. Amazon, which earns more than $68bn a year from ads on its own pages, blocked Muse days earlier.

Anthropic and OpenAI cut the price of their new AI models. They are the first releases since both backed a call to slow down. →

Anthropic's Opus 5.5 costs about 40% less to run than the model before it, and OpenAI's GPT-6 Sol and Luna cost half as much. Customers are suing the labs over their slowdown pact, record labels sued Suno again, and robotaxis grew wherever a permit allowed.

Texas stopped issuing every permit a data centre needs. California told data centres to publish what they use. →

On the same Monday, one US state shut both its doors on new data centres and another forced them to declare their power, water and land. In the markets, two companies that sell power to AI could not find buyers for their shares.

China is holding back humanoid robot share sales after Unitree fell 55%. Toyota says it will need 400,000 robots. →

China's securities regulator has quietly slowed approvals for humanoid robot companies that want to list, and is asking whether their sales to local-government ventures are real demand. In the same week Toyota put a number on the machines it wants, and Agility showed a humanoid that sits on the floor when a person walks near.

Apple's new Mac software has no switch to turn off its AI. It went out to every supported Mac on Monday. →

Apple shipped iOS 27 and macOS 27 Golden Gate with a rebuilt Siri, fixed a record 326 security flaws between them, and quietly dropped the setting that let Mac owners delete Apple Intelligence from their disk.

OpenAI's newest model now takes more business AI spending than Anthropic's. Anthropic is weighing a new model before its share sale. →

Two spending trackers put OpenAI's Astra ahead of Anthropic's Claude, six days after Anthropic's chief asked the industry to slow down, while many businesses say last year's cheaper models do most of the work.

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. →

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.

AI is not firing office workers. It is slowing their pay and their hiring. →

A study of 321 US jobs finds pay growing slower where AI can do the work, while the number of people employed has not changed.

SK Hynix is in talks to make memory chips inside Intel's Ohio plant. Production there was meant to start in 2025 and is now due in 2030. →

South Korea's SK Hynix, the world's second-largest maker of memory chips, is discussing leasing part of Intel's unfinished Ohio site. The US government can tax it for staying home and South Korea's government can rule the technology too sensitive to leave.

Microsoft wrote a rulebook for its own AI models. It does not cover the OpenAI and Anthropic models Microsoft also sells. →

Microsoft's draft AI code bans hidden reasoning and rejects rights for models, but reaches only the models it builds. Also: China's security minister names two US models, Korea's chipmakers refuse to prepay for power, and Windows fixes break sound and pasting.

Anthropic picked the Nasdaq for its share sale. In Asia on Monday, AI shares fell after its boss asked the industry to slow down. →

Anthropic, the company behind the Claude AI models, will list its shares on the Nasdaq, possibly in October, after filing at a value of $965 billion. On Monday SoftBank fell as much as 13.2% in Tokyo, after Anthropic's boss asked AI companies over the weekend to slow down.

Anthropic's chief says AI must slow down, and OpenAI agreed within a day. Neither said it would train any slower. →

Dario Amodei asked the industry to "pace the frontier" on Saturday and committed Anthropic to letting outside evaluators work inside the company. Sam Altman said OpenAI will do the same. Both want the US government to permit the safety talks that competition law would otherwise forbid.