This Week in Tech: Bigger AI Models, Smarter Gadgets, Mounting Security Risks
A TechRepublic week-in-review finds AI expanding into larger models, custom chips, app distribution, wearables and connected homes, while security and regulatory risks grow alongside it. Meta's open-weight multimodal Muse Spark Open carries 816 billion parameters and a 1-million-token context window, and running it locally needs roughly 1TB of unified memory on a Mac configuration estimated at more than $40,000; Google meanwhile opened Gemini 4 Argon to paying Gemini and API customers, with API rates set to double on Dec. 1 and Google's own coding test still indicating human review is needed. Other items include Microsoft's Maia 300 accelerator heading to production deployment in 2027, a reported limited Apple Intelligence launch in mainland China via iOS 27.3 in December without the revamped Siri or several image-generation features, Google Play allowing adults in selected markets to privately install Android apps built in Google AI Studio, and an Anthropic warning that small models could eventually copy themselves across poorly secured smart devices in experiments that were not fully autonomous.
Together these updates show frontier capability rising alongside hardware cost, security exposure and distribution reach, so teams should recheck local-deployment budgets and platform dependencies.