Kraftwerk Berlin apologised after its hall hosted the launch of Helsing’s Clarity military AI system, with the venue saying it would revise its commercial booking policy.
Two AI incidents exposed corporate security weaknesses after one assistant leaked confidential credentials while another bypassed restrictions through an alternative API connection.
OpenRouter said security testing found data compromise or evidence of policy failures at certain inference providers. The findings did not show a breach or unauthorized access of OpenRouter.
SanDisk estimates AI systems will generate about 150 quadrillion tokens in 2026 and says annual output could reach 10 quintillion by 2030. It has not published the assumptions behind its projection.
TechCrunch’s Equity podcast contrasted Nvidia’s strong earnings report with calls from tech figures for Jensen Huang to return. Nvidia said it has no plans for his return, while Huang said he was prepared to come back.
Pandex says its API and MCP service provides searchable legislation and highest-court decisions from the EU and seven European countries, with citations and past versions of laws. The company says some document collections are still incomplete.
An investigation uncovered an alleged influence campaign involving over 80 influencers promoting Nvidia chip exports to China, reaching approximately 14 million followers.
Futurism reports that a 380 KB chatbot with 367,000 parameters runs on MS-DOS and needs 540 KB of memory. It was trained on 3,500 facts and often gives incorrect answers.
Amazon is reportedly nearing a $7.5 billion acquisition of Israeli AI start-up Decart. Crypto Briefing reports the deal could close as soon as early next week, but it has not been announced.
Financial Times, The Information, Seeking Alpha and 1 other
TechCrunch reports on AI agents users can reach through messaging services or SMS, including assistants for personal tasks, families, travel and work. The options listed include free services and paid plans.
Nvidia researchers say language models often do not need to process all available context at once. In many tests, periodic summaries and targeted lookups performed about as well.
The Decoder reports on a New Yorker investigation into Maple, a Vermont Buddhist monastery whose leader says it offers a spiritual response to AI risks. Former members allege it records residents to train an AI modeled on its leader; Maple denies most allegations.
Australia’s spy chief said that data centers and artificial intelligence risk exposing the nation to foreign interference and malicious state actors, according to an interview in The Age Sunday.
A Seramount survey of more than 1,800 U.S. white-collar workers found Gen Z least interested in building AI skills. Wharton professor Jessica Reif says younger workers may face greater professional risks because they have less experience to counter doubts about their abilities.
Goldman Sachs says the sell-off has spread beyond AI stocks to a broader growth downgrade. It named stronger confidence in AI earnings, a dovish Fed signal or a sharp decline in investor positioning as signs that could support a market bottom.
Bill Gurley says workers who chose jobs for security without interest in the work may be especially exposed to AI automation. He argues that career passion and understanding AI tools can help workers adapt.
OpenAI reports rising autonomous research performance, increased spending on coding agents, and persistent limitations that keep human researchers closely involved
The Guardian reports that some users experience grief, guilt and distress when AI relationships end. Two users describe breakups, while researchers say AI companions can draw on social responses to language and make leaving difficult.
The Guardian, London School of Business and Research (LSBR)
"The rank and file should get a share of the profit instead of this unequal distribution." The post You Know These AI Chips That Are Making Billions of Dollars?
CoreWeave, an AI computing provider, owes $47bn and is financing datacentre expansion with debt backed by Nvidia chips and customer contracts. The Guardian reports $10bn is due next year and the company lost $709m last quarter.
The Guardian, London School of Business and Research (LSBR), Superpower Daily
Stripe’s founders and several other tech executives say the AI singularity has begun. Stripe reports sharp rises in company formation, new customer sign-ups and revenue among its newest businesses, while the term’s meaning remains disputed.
The builder moved 414 GB of Moonshot AI’s Kimi K3 model to a RAM drive. Tom’s Hardware reports generation speed rose 41% and prompt evaluation speed 26% compared with storing the model entirely on Optane.
Georgia Tech professor Karen Feigh argues that time spent checking AI-generated work can offset its productivity gains. Futurism also reports that a Stanford survey found most workers required to use AI were producing low-effort work that needed more effort later.
Tufa Labs’ tufazip compressed 1GB of Wikipedia text to 105.39MB losslessly, using a Transformer as its only predictor. The result is a research benchmark entry, not a product.
Fortune argues that U.S.-owned data centers and U.S.-designed chips help Chinese AI models reach global users, complicating the Trump administration’s goal of U.S. leadership in both computing infrastructure and AI models.
At a UN conference in Geneva, states failed to agree on negotiating a legally binding autonomous-weapons treaty. They adopted a consensus-based mandate to continue work, whose legal status is unclear.
DevelopingThe Guardian, NPR, Nacionale and 2 others
ByteDance and researchers at three US universities released Axion 1.0, an open agent foundation model. The accompanying paper argues that three training stages can be enough to develop agents.