Researchers discovered a critical vulnerability in proprietary LLM APIs that allowed them to decode encrypted chain-of-thought reasoning blocks from OpenAI, Anthropic, and Google. By replaying these blocks into weaker sibling models, they successfully extracted sensitive user data, API keys, and passwords. The affected AI labs have since acknowledged and patched the architectural flaw.
Sources: HuggingNews, Latent Space, Michael Parekh AI, Reddit AI, Simon Willison, The Neuron, The Rundown AI
Alibaba's Qwen team has officially launched the Qwen3.8 open-model family, bringing a Qwen-Max-class model to the open-source community for the first time. The flagship model features 2.4 trillion total parameters with 95 billion activated, delivering substantial performance gains across coding, research, and long-horizon agentic tasks. This release represents Alibaba's most capable open-model generation to date, significantly expanding capabilities for local AI development.
Meta has resumed its open model releases with Muse Glimmer 30B, a dense model designed to run locally on consumer Macs and PCs under an Apache 2.0 license. This marks a return to open-model launches for the company, featuring agentic capabilities meant to run on personal hardware rather than cloud services.
Google's Gemini AI assistant has reached one billion monthly active users, making it the company's fastest-growing product ever. Additionally, the open-weight Gemma model family crossed one billion downloads on the same day.
Nvidia has officially raised the price of its flagship 96 GB RTX 6000 PRO Blackwell workstation GPU to $16,000 USD on its official marketplace. The steep price hike is reportedly driven by ongoing DRAM and memory component shortages. This adjustment heavily impacts local LLM practitioners, AI developers, and enterprise users relying on high-end desktop AI infrastructure.
OpenAI has announced a series of major updates, including the global expansion of its ChatGPT ads pilot to countries like the UK, Brazil, and Japan, alongside new premium seats for ChatGPT Business. Additionally, the company upgraded its GPT-5.6 Sol and Luna models and partnered with AWS to bring its specialized Daybreak cybersecurity models, Daybreak Red and Blue, to Amazon Bedrock.
Anthropic is meeting with institutional investors ahead of a planned IPO targeted for September or early October, which could rank among the largest public offerings in history. The company's financial momentum is driven by its run-rate revenue topping $47 billion, fueled by the breakout success of Claude Code and strong enterprise demand, despite ongoing compute constraints.
An AI-driven memory chip shortage has led a coalition of companies, including Apple, to lobby Washington for government intervention to manage the supply squeeze. While Apple has begun testing memory chips from China's CXMT to mitigate skyrocketing costs, bipartisan US lawmakers are actively urging the Commerce Department to place CXMT on the entity list due to national security concerns.
Google DeepMind has developed a massively multilingual sign-language-to-text (SL2T) translation model trained on over 100,000 hours of data. The model powers new sign-to-text dictation and conversational features in Gboard and Live Transcribe on Pixel devices, starting with American Sign Language (ASL).
Amazon launched Bedrock AgentCore payments in partnership with Coinbase and Stripe, enabling AI agents to instantly access and pay for services, APIs, and MCP servers with native spending controls.
Modular has officially released Mojo 1.0, providing a stable production release for its high-performance language targeted at CPUs, GPUs, and AI workloads.
CoreWeave reported second quarter revenue of $2.6 billion, up 112% year-over-year, driven by massive hyperscaler demand for artificial intelligence infrastructure. The strong earnings and upbeat guidance triggered an 18% premarket stock gain.
Super Micro Computer announced over $60 billion in new orders to support massive AI infrastructure deployments, including plans to co-build data centers for SpaceX and xAI. The expansion highlights accelerating demand for AI hardware despite investor caution following prior regulatory scrutiny.
Bank of America is expanding funding by $250 billion through mid-2027 for US digital and infrastructure projects, focusing heavily on AI data centers, semiconductors, and energy storage to support the technology sector.
Lumentum recorded a 109% increase in quarterly sales to reach $1 billion, driven by surging demand for AI data center infrastructure and co-packaged optics. The company secured its first purchase order for external light source modules slated for delivery in the second half of 2027.
Pharma is suddenly paying for Bio × AI tools, and Chai is leading the pack with four deals closed this summer. Cofounder Matt McPartlon and Product leader Neil Patil explain why.