Fibocom has unveiled a USB-powered AI dongle powered by the Qualcomm QCS6490 octa-core system-on-chip, designed to bring local large language model (LLM) processing to standard laptops and network storage devices. The release targets the expanding edge AI sector by offering a plug-and-play solution for offline AI tasks, such as transcription and search, without relying on cloud infrastructure.

Warner Music Group and artificial intelligence startup Suno have announced a settlement to their copyright dispute, establishing a new licensing partnership. Under the agreement, Warner artists can voluntarily opt in to have their work used in Suno’s generative models. The deal coincides with Suno’s USD 250 million Series C funding round and its acquisition of the concert-discovery platform Songkick from Warner.

Following the release of GPT-5 and strategic hires from Doximity and Instagram, OpenAI is reportedly developing consumer-facing health tools. The initiatives include a potential personal health assistant and data aggregation service designed to serve ChatGPT’s 800 million weekly active users, moving the company beyond infrastructure into direct patient utility.

In their new book Rewiring Democracy, experts examine how AI is reshaping global politics, from Brazil’s judicial systems to an AI avatar engaging voters in Japan. The analysis highlights efficiency gains alongside significant risks of bias found in unregulated German election tools.

A new study from the University of California, Riverside introduces "Federated Carbon Intelligence," a software framework designed to optimize AI data centers. By balancing real-time hardware health with local power grid data, the system aims to reduce carbon emissions significantly while extending the operational life of server fleets.

The Federal Bureau of Investigation (FBI) has reported that cybercriminals have stolen more than USD 262 million from US bank accounts in 2025 through account takeover (ATO) fraud. With more than 5,100 victims reporting incidents this year, the average loss per case has exceeded USD 51,000. Authorities warn that attackers are increasingly leveraging generative AI and deepfake technology to bypass security measures and impersonate trusted contacts.

Microsoft is lowering the entry cost for its AI assistant with a new US$21 business tier and introducing Agent 365 to oversee autonomous workflows. Despite reporting 150 million monthly users, the tech giant faces ongoing scrutiny from IT leaders regarding integration complexity and return on investment, prompting a strategic shift toward multi-model support with Anthropic and enhanced enterprise governance controls.

The World Bank’s Digital Progress and Trends Report 2025 documents a surge in generative AI adoption in middle-income economies, which now account for over 40% of global ChatGPT traffic. However, significant disparities in data center infrastructure and digital skills persist between high- and low-income nations, raising concerns about the potential for an uneven economic impact.

Amid a global surge in demand for artificial intelligence literacy, IBM’s “AI Foundations for Everyone” specialization on Coursera offers a no-code pathway for non-technical professionals. This report examines the four-course curriculum, the context of Coursera's 10 million GenAI enrollments, and how the program compares to competitor offerings from Google and DeepLearning.AI.

OpenAI enters the browser market with Atlas, a Chromium-based tool for macOS featuring a persistent ChatGPT sidebar and autonomous Agent Mode. The release highlights a shift toward AI-mediated browsing but draws scrutiny from security researchers regarding prompt injection risks and data privacy.

Following the Wang Fuk Court fire on November 26, which affected seven towers and resulted in at least 128 casualties , TheDayAfterAI utilized the AI assistant Claude to examine the event through fire science principles. The resulting analysis addresses the public debate regarding the roles of bamboo scaffolding versus plastic safety nets , suggesting that a combination of factors, including the "chimney effect" and specific material properties, contributed to the disaster.

As the death toll rises to 128 in Hong Kong’s Wang Fuk Court fire, investigators focus on flammable bamboo scaffolding and faulty safety systems. Police and the ICAC have arrested 11 suspects connected to the estate’s ongoing renovation works, while search operations for approximately 200 missing residents continue.

Despite major 2025 upgrades from OpenAI and Google, Hong Kong users face restricted access to global chatbots due to regulatory friction and corporate geoblocking. This report examines the impact on local businesses and the city’s strategic pivot toward developing sovereign AI models like HKChat.

Google’s Gemini 2.5 Flash Image, codenamed "Nano Banana," has emerged as a market leader for generating legible text within images. Independent testing indicates the model achieves a 94% success rate on typography prompts, significantly higher than Midjourney’s 71%. This article analyzes the technical shifts enabling this accuracy, details the token-based pricing structure, and outlines the capabilities of the recently released Nano Banana Pro.

With over 1,900 AI avatars and rapid adoption in the education sector, Vidnoz AI has established itself as a major player in synthetic video creation. However, the platform’s pricing structure, defined by rigid monthly minute caps that do not roll over, is generating friction among freelancers and small businesses. As competitors move toward unlimited billing models, users are increasingly debating whether Vidnoz's expiring credit system remains viable for professional content workflows.

Autonomous agent platform Manus is receiving sustained criticism over its credit-based pricing, which users claim is unpredictable and too restrictive for serious development work. The report examines the platform's economics, the $75 million funding round led by Benchmark, its technical architecture using models like Claude 3.7 Sonnet, and how its variable cost model contrasts with the predictable subscription pricing of competitors like GitHub Copilot.

Anthropic releases Claude Opus 4.5 on Nov 24, 2025, featuring "Infinite" context compaction and a 200K token window. The flagship model scores 80.9% on SWE-bench Verified and drops API input pricing to $5 per million tokens, challenging GPT-5.1. Adopted under the ASL-3 safety standard, Anthropic holds a 32% enterprise market share.

Anthropic has released Claude Opus 4.5, a new frontier model engineered for complex coding and agentic workflows. Achieving a record 80.9% on the SWE-Bench Verified benchmark, the system outperforms key rivals while introducing variable "Effort" settings that can reduce token consumption by up to 76%. With API pricing lowered to US$5 per million input tokens, Opus 4.5 targets heavy enterprise adoption across major cloud platforms including AWS, Google Cloud, and Microsoft Azure.

Even Realities has released its second-generation smart glasses, the Even G2, priced at US599, alongside the US249 R1 smart ring. Weighing just 36g, the camera-free G2 utilises a monochrome micro-LED display for privacy-conscious users, offering tools like live translation and teleprompting. A launch promotion offers the ring at 50 per cent off when bundled with the glasses through 26 December 2025.

A review of research published between mid-2024 and late-2025 highlights persistent political and social biases in large language models. Studies from Stanford and other institutions show how training data and safety filters shape model outputs, potentially influencing public opinion and automated decision-making in sectors like recruitment.