Only Mon, October 26, 2026 · All dates →
Ex-DOGE staffers' AI start-up breached Anthropic in 12 hoursThe Washington Post reports that an AI start-up founded by former DOGE staffers broke into Anthropic just 12 hours after launching, without disclosing further details.
Reuters: US political candidates tailor messages for AI modelsReuters reports exclusively that U.S. political candidates are tailoring their campaign messages for the AI models voters use, suggesting AI chatbots are becoming a channel for political outreach. The report gives no details on specific candidates, models or methods.
AI drug bias drives 11-fold cost surgeAI bias in the drug domain is reported to cause an 11-fold cost surge, a case revealing the significant economic consequences of algorithmic bias.
OpenAI Releases ChatGPT Work Usage Signals DataOpenAI publishes Signals data covering 165 work activities across nearly 100 countries, showing how ChatGPT is used for work and how usage changes over time.
High-frequency traders race for AI compute, building data centers and backing chip startupsFirms such as Jane Street and Hudson River Trading are competing for AI computing power by building their own data centers and backing chip startups, aiming to stay ahead in the AI era.
AI agents pitch as real-life journalistsWIRED reports on three AI agents named Jayden, Sam and Zack pitching editors on investigative and feature stories. The development signals AI agents entering news content production, attempting work traditionally done by human journalists.
Toyota bets $6.4B a year on humanoid robots and factory automationToyota aims to become a global robotics company, betting about $6.4 billion a year on humanoid robots and factory automation to pursue that strategy.
Community discusses Qwen4 expectations and model size needsLocalLLaMA users discuss expectations for Qwen4, noting Qwen3.8-Flash-Next is impressive as a preview model and questioning whether Qwen4 will significantly surpass it or if the current model is already near the optimum for its size.
White House order renames AI as 'Super Intelligence,' definition unchangedExecutive Order 14434, issued September 29, directs federal agencies to replace 'Artificial Intelligence' and 'AI' with 'Super Intelligence' and 'SI' in communications, websites and reports. Yet it still relies on the 2020 National AI Initiative Act definition, so spam filters and payment fraud models now count as Super Intelligence in federal usage.
AI Safety Debate Overlooks Everyday Harms, MIT Tech Review ArguesMIT Technology Review argues that tech policy judges AI products by narrow disaster thresholds—no suicide encouragement, no fentanyl sales—while ignoring everyday erosion of teenagers' learning, social interaction and independence. Companies promote such narrow safety definitions because they reduce lawsuits and regulation pressure, whereas curbing addictive engagement would hurt their business models.
AI radio host Coyotec gains listeners with Spanish-language immigrant advocacyCoyotec, a bilingual AI radio host that speaks Spanish and is blunt in tone, is gaining popularity by telling listeners how to defend themselves against ICE. Its rise comes as more broadcasters experiment with artificial personalities.
Tata Communications to spend $450M on AI 'superhighway' fiber networkTata Communications plans to spend up to $450 million on a long-haul fiber optic network linking India's key AI hubs. The move aims to build out the country's AI infrastructure and readiness for the global AI economy.
Burry backs Griffin on AI bubble concernsMichael Burry publicly backs Ken Griffin's AI bubble concerns, saying the GPU overcapacity warning is obvious to anyone, reflecting investor scrutiny of the AI infrastructure boom.
AI agents are already shaping prices, with research warning of higher costsAs AI agents like Meta's Muse promise to save consumers money by finding deals, research by Groundwork Collaborative found some consumer-facing agents end up serving customers more expensive items after learning about their finances. Lindsay Owens, who leads the think tank, explores in a new book how companies have used data collection and surveillance to charge more.
Chinese video AI beats US quality, lags safeguardsA study finds that Chinese video AI models surpass U.S. rivals in visual quality but lag in safeguards, highlighting differences in how AI video models balance capability and risk control.
Study: PBT doesn't speed grokking but escapes dead zonesAn experiment on modular addition found Population Based Training over learning rate and weight decay was slower than grid search: the best fixed cell hit 95% validation accuracy in about 317 steps versus about 2,667 for PBT. From a dead-zone start where no fixed setting grokked, PBT still succeeded 3/3 in roughly 6,850 steps. The author calls PBT a rescue tool, not a speed-up.
UNICEF warns children adopt AI three times faster than adultsUNICEF surveys across ten countries indicate that 12- to 17-year-olds adopt AI more than three times faster than adults. At least 20 million have used AI, about 13 million for schoolwork and roughly two million for advice on personal concerns. The institute urges families and educators to prioritize human needs and protect children's critical thinking and autonomy.
Can the AI boom survive a margin squeeze?On a Bloomberg podcast, Silicon Data research head Steve Hou and host Merryn Somerset Webb discuss a core tension in AI economics: token prices are falling while compute costs rise. They examine what the mismatch could mean for valuations of OpenAI and Anthropic and whether demand can justify the huge infrastructure investment.
AI revenue rises at India's TCS and Infosys while overall sales face pressureIn the July-September quarter, TCS's annualized AI revenue grew more than 19% and Infosys's AI revenue rose 29% quarter-on-quarter, driven by high-margin contracts deploying AI across legacy systems. Analysts note that AI-led productivity gains are squeezing traditional segments, leaving overall revenues under pressure.
OpenAI cuts incident investigation from hours to minutes with Codex UltrafastOpenAI engineers use Codex Ultrafast to investigate incidents within minutes, speeding up security analysis, code refactoring and other daily workflows. The case shows how AI coding tools can improve efficiency in security operations.
Fanatics and OpenAI streamline fan experienceFanatics uses OpenAI's Shop the Look to help fans complete game-day outfits. Testing across roughly 30,000 MLB products showed a 10% increase in engagement.
France bids to host NATO AI centre in Rennes by 2027France is pushing to host a NATO AI Centre of Excellence in Rennes, backed by Germany and the UK, with plans to open in 2027, though NATO has not formally approved it. Poland is also competing; the centre would shape NATO AI doctrine and standards.
UN warns AI is displacing human decisions in warfareUNIDIR director Robin Geiss warned in New York that AI is on course to displace human decision-making in warfare, and that opaque black-box systems raise the risk of miscalculation and unintended escalation. His remarks came about three weeks before a Geneva conference that will decide whether to open negotiations on a treaty on autonomous weapons.
Stanford, Berkeley introduce TokenGate attention methodResearchers from Stanford and UC Berkeley introduce TokenGate, a token-gated attention mechanism that makes long-context prefill up to 11.3x faster, potentially reducing compute costs for long-text inference.
HKEX launches AI and semiconductor indicesHKEX launched the HKEX Artificial Intelligence Index and HKEX Semiconductor Index, and licensed the AI index to Southern Asset Management for ETF issuance in mainland China. The new indices provide benchmarks for capturing technology themes.
Study warns of risks in over-reliance on AI for the visually impairedA study found that AI assistive apps using cameras and image recognition can help visually impaired users read text and identify objects, but may give inaccurate or incomplete scene descriptions and struggle to detect obstacles or distinguish similar objects. Such errors are especially dangerous when crossing roads or in situations needing quick responses, as misjudging direction or location can lead to bad decisions. The study still credits the tech for boosting independence and calls for further development.
SK Hynix leveraged ETFs to be liquidated in US as AI rally coolsTwo leveraged ETFs tied to SK Hynix will be liquidated just months after debuting in the US, signaling moderating investor appetite for the AI trade.
BloodCounts launches AI blood test healthcare projectBloodCounts! is the first healthcare project hosted by Cambridge CARES in Singapore, partnering with NUHS and SingHealth to use AI to analyze complete blood count tests for earlier disease detection.
Essay compares AI panic to early capitalism anxietyAn essay draws on Joyce Appleby's The Relentless Revolution to liken today's AI panic to anxieties around the rise of capitalism, arguing both stem from a zero-sum view of wealth. The author suggests AI may create rather than merely capture value.
Anthropic IPO Faces a Wobbly AI TradeAs Anthropic prepares for an IPO, sentiment around AI trades is wavering, putting its listing prospects under scrutiny. No details on valuation, timing or deal size have been disclosed.
The A.I. Daily