TL;DR

U.S. judge backs fair use for Anthropic AI training; Google, Nvidia, HPE launch major AI tools; Scale AI data leak.

Highlights

• U.S. judge rules Anthropic’s use of legally purchased books for AI training is fair use, but piracy claims over 7M copied titles proceed to trial—first substantive federal decision on AI copyright 1.
• Bipartisan U.S. bill would bar federal agencies from using Chinese, Russian, Iranian, and North Korean AI models; DeepSeek specifically cited 2.
• Google launches open-source Gemini CLI, free for most developers, with up to 1,000 daily requests and security features 3.
• Google DeepMind debuts Gemini Robotics On-Device, a VLA model running locally on robots, adapting with 50–100 demonstrations, and SDK support 4.
• Google DeepMind introduces AlphaGenome, an AI model for analyzing up to 1M DNA bases, available via API for non-commercial research 5.
• HPE and Nvidia launch pre-integrated “AI Factory” stack with Blackwell GPUs, targeting enterprise and sovereign AI deployments 6.
• Rubrik acquires AI startup Predibase (>$100M) to enhance agentic AI and secure ML deployment for enterprise clients 7.
• Samsung unveils 3nm Exynos 2500 chip with 39% improved on-device AI performance for upcoming foldables 8.
• Scale AI exposes confidential client and contractor data via unsecured Google Docs, raising cybersecurity concerns in AI data handling 9.
• Goldman Sachs deploys generative-AI assistant firmwide, joining other major banks in rolling out internal LLM tools 10.
• OpenAI develops unreleased ChatGPT productivity features to rival Google Workspace and Microsoft Office , intensifying competition with partners 11.
• OpenAI and Microsoft CEOs discuss partnership tensions as some enterprise clients shift from Copilot to ChatGPT; OpenAI hints at a powerful open-source local model 12.
• Nvidia reclaims top market cap from Microsoft ; Broadcom upgraded by HSBC on strong AI chip demand and revenue outlook 1314.
• Google and Warp launch agentic AI coding tools, reflecting increased demand for autonomous developer workflows and multi-agent systems 15.

Commentary

The U.S. court’s mixed ruling in the Anthropic copyright case is the first to substantively address fair use in AI training, drawing a line between legally acquired and pirated data 1. This will inform risk management and data sourcing strategies for model developers facing ongoing litigation. Companies relying on broad web scrapes or unlicensed datasets should expect heightened legal scrutiny and may need to accelerate investment in licensed or proprietary data pipelines 1.

Regulatory momentum continues with the proposed “No Adversarial AI Act,” signaling increased barriers for AI models originating from China and other adversarial states in U.S. government procurement 2. DeepSeek’s inclusion and concerns over Nvidia chip access highlight geopolitical sensitivities around both AI software and hardware 2. Vendors and integrators serving public sector clients should closely monitor compliance requirements and prepare for more rigorous provenance checks 2.

On the product front, Google is expanding developer access and edge capabilities with the open-source Gemini CLI 3 and DeepMind’s on-device robotics model 4, while AlphaGenome’s release underscores AI’s growing role in genomics research 5. HPE and Nvidia ’s AI Factory stack 6, along with Samsung’s 3nm Exynos 2500 8, point to intensifying competition in AI infrastructure—both in the cloud and at the edge.

Enterprise adoption is accelerating: Rubrik’s Predibase acquisition targets secure, scalable agentic AI for data-centric clients 7, while Goldman Sachs’ full-firm AI assistant rollout reflects the normalization of LLMs in financial services 10. Meanwhile, OpenAI’s unreleased productivity tools 11 and its evolving relationship with Microsoft 12 signal rising competition and shifting alliances in the productivity AI space. The agentic AI trend is further reinforced by new launches from Google and Warp 15, and growing demand for workforce upskilling in multi-agent systems 15.

Security remains a weak link, as highlighted by Scale AI’s exposure of sensitive data through public Google Docs 9. As AI adoption scales, robust data governance and operational security must become standard practice, especially for vendors handling client IP or personal information 9.