Show only each day's banner story

Nvidia launches investigation: Jensen Huang may have been targeted by foreign intelligence

Security at a key SI chipmaker has become a corporate governance issue: the board is examining the circumstances surrounding the CEO, although it currently sees no reason to believe company secrets were compromised.

A committee of independent directors working with outside counsel will oversee the investigation. The board said it had no reason to believe that US national security or Nvidia secrets had been compromised, while Huang supported the review and promised to cooperate. According to The Guardian and TechCrunch, Huang acknowledged in a message to employees that he had a romantic relationship with Alice Huang, a former executive at Singapore-based Megaspeed, and apologized to his family and the company. The New York Times had previously reported, citing unnamed sources, that US officials believed she was linked to China's Ministry of State Security. According to The Guardian, the US congressional foreign affairs and intelligence committees asked Huang to attend a closed-door briefing next week on protecting Nvidia from foreign espionage.

SI development increasingly depends not only on chips but also on dedicated energy supplies: a lab can take over a place in the queue for generators and start computing without waiting years for a grid connection.

Anthropic buys commitments for 1.7 GW of gas generators for $3.4 billion

SI development increasingly depends not only on chips but also on dedicated energy supplies: a lab can take over a place in the queue for generators and start computing without waiting years for a grid connection.

According to the Financial Times, Anthropic acquired orders from OnePWR Solutions in late September for large Caterpillar gas generators with a combined capacity of 1.7 GW, together with installation work. The deal was worth $3.4 billion; the FT calls it the largest known single purchase of power-generation equipment by an SI lab. OnePWR ordered the generators about two years ago but needed financing to complete the purchase. Anthropic did not acquire OnePWR itself, and none of the supplier's employees moved to Anthropic. Anthropic and OnePWR did not respond to requests from Reuters and the FT, and Reuters could not independently verify the report.

Siri could move beyond voice commands to independently handling extended requests and actions in third-party apps, bringing it closer to competing agents.

Apple explores turning Siri into an autonomous SI agent

Siri could move beyond voice commands to independently handling extended requests and actions in third-party apps, bringing it closer to competing agents.

According to Bloomberg, Apple CEO John Ternus is demanding a more ambitious SI strategy from software chief Craig Federighi. The company wants to turn Siri into an autonomous agent that can track a goal, determine the next steps and continue working without a separate user command for each action. According to The Verge's account, upcoming updates will give Siri more precise control over Apple's own apps, such as cropping and rotating photos and organizing Safari tabs. These capabilities are expected in iOS 27.3, while actions in third-party apps, such as ordering food through Grubhub or booking through OpenTable, are planned for later in the spring. A more autonomous version of Siri could arrive as early as next year.

Amazon is returning to the idea of a home robot after Astro's limited launch, drawing on more capable SI models and the team at humanoid developer Fauna.

Amazon assembles new team to develop a home robot

Amazon is returning to the idea of a home robot after Astro's limited launch, drawing on more capable SI models and the team at humanoid developer Fauna.

CNBC found Amazon job listings describing a new robot intended to navigate changing home environments autonomously and operate alongside family members. The project sits within the Personal Robotics Group, which includes members of the Astro team and Fauna Robotics, acquired in March and creator of the roughly 107 cm-tall Sprout humanoid. The listings also cover customer support, on-site repairs, and designing everyday interactions with the robot. Amazon spokesperson Brad Glasser confirmed the company's intention to develop personal robotics and said it continues to support Astro owners.

Sources 1

The independence of US government SI evaluations is now in question: trust in these assessments depends on who sets CAISSI's priorities and controls publication of its results.

Staff at US center CAISSI asked to report informally to the White House

The independence of US government SI evaluations is now in question: trust in these assessments depends on who sets CAISSI's priorities and controls publication of its results.

According to an Associated Press report relayed by Unite.AI, two former government officials described the request to staff at the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) within the US Commerce Department. The agency's sources linked it to concerns about the independence of government safety evaluations of frontier models; no formal reorganization has been announced. CAISSI assesses cyber, biological and chemical risks and works with developers under voluntary agreements. According to Unite.AI, its annual budget is $10 million, it currently has no permanent director, and Congress has been asking since August about the cessation of report publications. Supporters of closer White House ties believe they could raise the division's profile and help secure funding; critics fear political influence over evaluation findings. The White House did not respond to AP's specific questions about CAISSI.

SI helps people draft formal complaints quickly, but also shifts the burden of reading and checking long texts onto public-sector employees.

UK public bodies report a flood of complaints and requests created with SI

SI helps people draft formal complaints quickly, but also shifts the burden of reading and checking long texts onto public-sector employees.

The Guardian described an increase in the volume and length of correspondence sent to UK local authorities, courts, schools, and MPs' offices. According to Deborah Evans, chief executive of Lawyers in Local Government, most complaints are now about ten times longer than the previous two pages, and one text she knows of runs to about 200 pages. The ICO noted that such requests can be repetitive, include unrelated comments and misdirected demands, but staff still have to answer them. Transport for London received a record 4,769 freedom of information requests in 2025/26, 5% above the previous peak; it also reported an increase in appeals that appeared to be generated with SI. Some councils are limiting complaint lengths or testing SI to summarise them.

Voice SI is moving closer to natural conversation: it can keep talking while searching for information instead of falling silent while processing a request.

Shizuku AI unveils Japanese voice model that listens and speaks simultaneously

Voice SI is moving closer to natural conversation: it can keep talking while searching for information instead of falling silent while processing a request.

Shizuku-Duplex processes dialogue in 80 ms steps: at each step, the model receives the other person's speech and generates its own text and audio tokens. This allows it to produce backchannels, choose pauses and accept interruptions without waiting for the speaker to finish. The system is split into a live interaction model and a background model. The latter calls tools asynchronously, such as web search, while the former continues the conversation. The developers also added context summarization approximately every five minutes for long conversations. In Shizuku AI's own test, the model produced the most natural dialogue among six systems evaluated and ranked behind only GPT-Live for response quality among full-duplex models. The median pause before a response was 0.57 seconds, compared with 0.53 seconds for GPT-Live; measurements for these two models excluded network latency.

Sources 1

The UK is choosing to expand computing capacity while tying government support to jobs, investment in local communities and closer oversight.

UK to continue building data centers for SI as essential infrastructure

The UK is choosing to expand computing capacity while tying government support to jobs, investment in local communities and closer oversight.

UK Growth Minister Dan Tomlinson told the Financial Times that wider use of SI across the economy would require new data centers. He said projects would need to invest in local communities and create jobs to receive government support; the government also intends to strengthen oversight of construction. The UK Compute Roadmap calls for data center capacity of at least 6 GW by 2030. Since 2025, the AI Growth Zones program has made it easier to approve such facilities. The statement came amid protests by local residents against individual sites and restrictions adopted in local votes in the United States.

Personalised memory makes an SI assistant more useful, but personal details require the same caution as with other services that store user data.

Russian Interior Ministry warns of risks from ChatGPT's personalised memory

Personalised memory makes an SI assistant more useful, but personal details require the same caution as with other services that store user data.

The Russian Interior Ministry's unit for combating the unlawful use of information technology warned that details about work, health, income, and personal experiences can be stored in ChatGPT's memory and form a digital dossier. In the ministry's assessment, this collection of data creates a risk of identifying users even when they use pseudonyms, while false instructions in processed documents can manipulate the service's memory. The ministry recommended checking memory and personalisation settings, not uploading confidential documents, anonymising requests, and using temporary chats for personal matters.

Russia now has serial production of another data-centre infrastructure component needed to house and modernise computing capacity.

Technopolis Moscow launches production of modular power cabinets for data centres

Russia now has serial production of another data-centre infrastructure component needed to house and modernise computing capacity.

System Electric has begun serial production in the special economic zone of cabinets that distribute electrical power in data-centre machine rooms and server rooms. The range includes single-phase and three-phase 16 A and 32 A modules that can be adapted to different loads. Moscow Deputy Mayor Maksim Liksutov linked the launch to plans to build and modernise Russian data centres. The company's existing facility also produces three-phase uninterruptible power supplies and houses an engineering and service centre.

Sources 1

The service-robot maker intends to expand into humanoids and bring potential manufacturing closer to its European customers.

Pudu Robotics plans to develop humanoids and considers manufacturing in Europe

The service-robot maker intends to expand into humanoids and bring potential manufacturing closer to its European customers.

According to the South China Morning Post's feed description, Pudu Robotics founder and CEO Felix Zhang Tao called humanoid robots the company's next direction and a long-term area of competition with Tesla. Pudu, which makes 80% of its sales abroad, is also considering manufacturing in Europe, its largest overseas market.

Sources 1

As models gain biological capabilities, the decision to publish open weights increasingly depends on risk testing, not just model quality.

Tencent chief scientist warns frontier models are approaching biological risks

As models gain biological capabilities, the decision to publish open weights increasingly depends on risk testing, not just model quality.

According to the South China Morning Post, relayed by The Next Web, Yao Shunyu told a science event in Hong Kong that models' biological capabilities could approach a threshold of practical harm, as has already happened in cybersecurity. He urged Chinese companies to take safety seriously when releasing open models and said biological risks could be harder to manage than cyber risks. The speech followed the launch of the Hy4 model on November 4. Tencent said it had excluded material on exploiting vulnerabilities from the model's training and used defensive security data instead; according to the company's internal assessment, Hy4 has basic biology knowledge but is weaker than leading models at designing experiments. Yao acknowledged Anthropic's arguments for keeping frontier models closed but proposed combining open and closed development because open models can also help defenders.

Australia must choose between SI companies' demands for easier access to creative works and a mandatory licensing model that the music industry says already works.

IFPI urges Australia not to weaken copyright law for SI training

Australia must choose between SI companies' demands for easier access to creative works and a mandatory licensing model that the music industry says already works.

Victoria Oakley, head of the International Federation of the Phonographic Industry, told The Guardian she opposed exceptions that would allow models to train on music without licenses. She linked potential concessions to a risk to creators' incomes and noted that record companies had already signed dozens of licensing agreements with smaller SI companies. The Australian government is considering copyright changes alongside other SI measures. Major laboratories describe legal certainty as a condition for investment; Anthropic has proposed a system in which material could be used unless the rights holder explicitly objected. Prime Minister Anthony Albanese promised strong protection for creators, while Attorney-General Michelle Rowland's office said any changes must give creators meaningful control and fair compensation.