Kimi K3 and follow-on models rattle US valuations
New Chinese models match or exceed frontier performance on key benchmarks while remaining openly available. This accelerates capability diffusion and pressures closed-model pricing and margins.
Read on Digital JournalHalluSquatting attack tricks AI agents into code execution
Adversarial prompts cause LLM agents to fetch and run malicious packages from public repositories. The technique exploits hallucinations in tool-use chains rather than traditional supply-chain compromise.
Read on The Hacker NewsDiscord AI moderators flag thousands of benign images
Automated moderation systems over-blocked safe content at scale, forcing manual review and policy rollback. The incident highlights limits of current vision classifiers in high-volume user platforms.
Read on TechCrunchAMD outlines new AI silicon and software roadmap
AMD details next-generation accelerators and a software stack aimed at training and inference clusters. The move positions the company against Nvidia in both data-center and edge AI workloads.
Read on Tom's HardwareAI designs QLED materials with 2x efficiency and 40x lifetime
Researchers used generative models to optimize quantum-dot LED recipes, yielding measurable gains in efficiency and durability. The result shortens iteration cycles for display and lighting hardware.
Read on EurekAlertWhite House directs access rules for frontier AI models
Administration guidance now shapes which organizations can obtain the latest model weights and APIs. The shift moves allocation decisions from vendors toward government-defined criteria.
Read on CNBCData-center opposition spreads across US states
Local protests against new hyperscale facilities are expanding nationally over power and land use. Operators face longer permitting timelines and potential capacity constraints in key regions.
Read on U.S. NewsXi calls for open AI standards in international push
China's leadership advocates broader model and data sharing to set global norms. The stance contrasts with US export controls and may influence standards bodies and allied policy.
Read on Tech StartupsChina's data scale emerges as durable AI advantage
Domestic data volume and regulatory access give Chinese labs training sets that remain hard to replicate elsewhere. This structural edge persists even as model architectures converge.
Read on SCMPCurated daily by Kingsmen Consultancy. Each evening the lead story gets the full plain-English treatment in our news section.