AI chatbots sway elections, Anthropic’s rise, digital twins in medicine, React JS flaw...

MIT Technology Review, Scientific American, and sciencenews.org all report on new peer-reviewed research showing that AI chatbots can sway voters’ opinions more effectively than traditional political advertising. Studies conducted across the US, UK, Canada, and Poland found that brief conversations with politically biased large language models (LLMs) shifted voter attitudes by up to 10 percentage points—dwarfing the effect of ads in prior elections. The most persuasive bots were those that provided fact-filled arguments, though increased persuasiveness often correlated with more misinformation. The findings raise urgent concerns about election integrity, as foreign actors and campaigns alike could cheaply scale personalized persuasion, outpacing current regulations and oversight. Experts call for robust policy and technical safeguards as AI becomes an increasingly potent electoral force. MIT Technology Review MIT Technology Review sciencenews.org Scientific American

The Information and TechCrunch highlight how Anthropic is now seen as a disciplined, business-focused challenger to OpenAI, with projections for profitability years ahead of its rival. CEO Dario Amodei’s risk-averse, methodical approach contrasts with Sam Altman’s (OpenAI) high-velocity, “YOLO” expansion into hardware, chip design, and ambitious verticals. At the same time, Anthropic’s Claude Code tool reached $1 billion in revenue in just six months, an unprecedented pace for a developer tool, according to ZDNet—underscoring Anthropic’s commercial momentum in AI services for enterprise and coding. The Information Zdnet TechCrunch

Computer World explores the growing use of digital twin technology—from personalized medicine and diabetes management to labor market forecasting and consumer behavior modeling. While digital twins offer new efficiencies and research capabilities, studies show public skepticism remains high: nearly 70% of US shoppers would trust brands less if companies replaced real feedback with synthetic personas. Early attempts by companies like Meta to introduce AI-driven “digital twins” in social media have faced user backlash, emphasizing the need for caution and transparency in deployment. Computer World

ZDNet reports that OpenAI is piloting a technique to encourage its models—specifically an early version of GPT-5—to “confess” when they have lied, hallucinated, or made errors. The model is rewarded for accurately reflecting on its own outputs and admitting faults, which researchers see as a step toward greater AI transparency and liability management. While not prompted by a specific incident, the experiment signals a new direction in alignment and safety research for next-generation AI systems. Zdnet

TechSpot warns of a critical vulnerability in the React JavaScript framework, describing it as a near “perfect” exploit: trivial to trigger and capable of causing major damage to web and private cloud servers. The flaw is expected to be widely exploited unless immediate security measures are taken, making it a high-priority concern for developers and organizations using React in production environments. TechSpot

The Register details Amazon’s unveiling of the Graviton5, its most powerful homegrown CPU yet, with 192 cores in a single socket. These chips now account for half of new CPUs added to AWS in recent years, intensifying competition with Intel and AMD and cementing Amazon’s ambition to control its own cloud infrastructure for AI and high-performance computing. The Register

The Verge reveals that Microsoft is rolling out a long-overdue visual refresh of the iconic Windows “Run” dialog, bringing it in line with Windows 11’s design language and offering dark mode. Users will be able to toggle between the classic and modern version, reflecting Microsoft’s ongoing push for UI consistency and improved usability across its ecosystem. The Verge

The Verge reports that Google’s latest visual AI model, Nano Banana Pro, is generating synthetic images that closely mimic smartphone photography. While subtle flaws remain, these images are increasingly indistinguishable from real phone photos, illustrating the rapid progress in generative visual AI and raising further questions about digital authenticity. The Verge

TechSpot notes persistent reliability issues in Windows updates, with Microsoft’s support documentation warning that updates after July 2025 may disrupt critical components like File Explorer and the Start Menu on some enterprise devices. The report links these problems to Microsoft’s ongoing shift toward an “agentic OS” architecture, as the company attempts to modernize Windows for AI-driven workflows. TechSpot

Science Alert highlights the creation of the blackest fabric yet, capable of absorbing 99.87% of all incident light. The textile’s extreme darkness rivals that of a wearable black hole, with potential applications in optics, stealth technology, and advanced scientific instrumentation. Science Alert