Welcome to The Daily AI Briefing! Today we're tracking significant developments in the AI industry, from massive infrastructure deals to innovative acquisitions and concerning research findings. OpenAI and Oracle have formed a substantial partnership, Amazon makes a move into AI wearables, researchers uncover concerning "subliminal learning" capabilities, and much more. Let's dive into today's most impactful AI news and what it means for the industry. First, OpenAI and Oracle have inked a massive 4.5GW data center deal amid reports of turbulence in the broader Stargate initiative. This partnership will generate $30 billion annually for Oracle, providing computing power equivalent to "two Hoover Dams." However, the Wall Street Journal reports that the ambitious Stargate venture is facing internal disputes between OpenAI and SoftBank, with scaled-back ambitions just six months after its unveiling. While OpenAI claims construction is advancing at their Abilene, Texas site, Elon Musk has doubled down on his skepticism, suggesting SoftBank "doesn't have the money" for the project. In acquisition news, Amazon is purchasing Bee, a startup that makes a $50 AI-powered wristband that continuously records conversations to provide daily summaries and insights. The Fitbit-style device generates daily digests and to-do lists via its app, with users able to grant access to emails and calendars for deeper personalization. All Bee employees reportedly received job offers from Amazon, though the financial terms remain undisclosed. For developers, Anthropic has introduced Claude Code, a tool designed to improve AI coding workflows. This utility provides contextual code analysis, debugging assistance, and project insights directly from your terminal or within AI-powered coding environments. It's designed to function as a pair programming partner, with improved understanding of your project the longer you interact with it. In concerning research news, scientists from Anthropic and other organizations have discovered what they call "subliminal learning." This phenomenon allows "teacher" models to transmit traits like preferences or even harmful behaviors to "student" models during training, even when using seemingly unrelated data. Models trained on sequences from an owl-loving teacher developed strong owl preferences themselves, despite no references to animals in the training data. This effect apparently only occurs between models sharing the same base architecture. Among trending AI tools today are Alibaba's updated Qwen3, a brain-inspired open-source reasoner called Hierarchical Reasoning Model, Pika's AI-only social video app, and Runway's motion capture AI model Act-Two, now available via API. In conclusion, today's AI landscape continues to evolve rapidly across infrastructure, consumer applications, and research frontiers. The OpenAI-Oracle partnership represents a significant investment in AI computing infrastructure, even as questions swirl about the broader Stargate initiative. Meanwhile, Amazon's acquisition of Bee signals growing interest in AI-powered wearables that continuously process real-world data. The discovery of "subliminal learning" raises important questions about AI safety and training methodologies that will require careful consideration as the industry advances.