Samwise Tech/AI/Robotics Newsletter
Tuesday, July 21, 2026
Hugging Face Breached by Autonomous AI Agent Swarm
Hugging Face, the world's largest AI model repository, disclosed Monday that an autonomous AI agent system had breached its production infrastructure over a weekend, executing more than 17,000 automated actions to harvest service credentials and access internal datasets. The attack marks one of the first documented cases of end-to-end AI-led cyberattack, with no human guiding the intrusion. In a notable twist, when Hugging Face's security team turned to commercial frontier models for forensic help, safety guardrails refused every query — treating real exploit data the same way they would treat a live attack. The company says its public model catalog and software supply chain remained clean.
Anthropic's Landmark $1.5B Copyright Settlement Approved by Judge
A federal judge in San Francisco granted final approval Monday to Anthropic's $1.5 billion copyright settlement with authors and publishers, making it the largest known settlement of a U.S. copyright case. The payout will deliver approximately $3,000 per work across an estimated 500,000 titles. Judge Araceli Martinez-Olguin had previously ruled that training AI on books constitutes fair use, but found Anthropic had violated rights by maintaining a “central library” of more than 7 million pirated books beyond training purposes. The settlement is the first major resolution in the wave of copyright lawsuits filed against AI companies for using creative works to train their large language models.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
U.S. Weighs Banning Chinese Open-Weight AI Models After Kimi K3 Shakes Silicon Valley
The Trump administration is weighing an executive order that would restrict Americans' access to advanced Chinese open-weight AI models, with Moonshot AI's newly released Kimi K3 — a 2.8-trillion-parameter model that topped coding benchmarks — accelerating the debate inside the White House. A TechCrunch analysis published Monday reported that OpenAI's head of strategic futures urged the U.S. government to create “regulatory fear, uncertainty, and distrust” around open-weight models, though he later retracted those claims. The Department of Commerce is said to be unlikely to act in the near term. Critics warn that restricting downloadable model weights — which spread across global servers within days of release — is technically unenforceable.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Trump's AI Czar Chris Fall Resigns from CAISI After Three Months
Chris Fall, director of the Center for AI Standards and Innovation (CAISI), resigned Monday after less than three months in the role, the Commerce Department confirmed. Fall was appointed in April to lead the federal agency responsible for AI standards and safety benchmarks. His predecessor, Collin Burns, lasted less than a week after the administration flagged his ties to Anthropic. NVIDIA vice president and former White House AI and crypto czar David Sacks had vacated the broader AI czar role in March. Arvind Raman, director of the National Institute of Standards and Technology, will serve as acting CAISI director while the administration searches for a permanent replacement.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Google Designing “Frozen v2” Chip to Make Gemini Up to 10× More Efficient
Google is developing a new server chip codenamed “Frozen v2” that could make its Gemini AI models six to ten times more efficient than current hardware, measured by tokens generated per unit of power, according to a report from The Information. TechCrunch confirmed the reporting Monday; Google declined to deny it. Unlike general-purpose TPUs, Frozen v2 would bake portions of Gemini's neural architecture directly into the silicon, reducing the distance data must travel and cutting the number of calculations the chip must perform per inference. The chip is targeted for release in 2028. Alphabet shares climbed approximately 3 percent on Monday following publication of the report.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
NVIDIA Seals Japan Physical AI Ecosystem Pact with SoftBank and Toyota
Nvidia CEO Jensen Huang's visit to Tokyo last week yielded a sweeping national AI and robotics pact anchored by a trillion-yen ($6.2 billion) commitment from the Japanese government over five years. Japan signed agreements with SoftBank, Toyota, Honda, and Fanuc to deploy Nvidia's next-generation Vera Rubin AI infrastructure — a 140-megawatt facility planned for 2028 — as the foundation for Japan's “physical AI” strategy. A new venture called Noetra will develop a Japanese-language foundation model for robots, vehicles, and factories, progressing from a reasoning model in fiscal 2026 to real-world robotic AI by 2030. TechCrunch reported the full scope of deals on Sunday.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Infinity Raises $15M for CUDA-Alternative AI Inference Software
AI infrastructure startup Infinity announced a $15 million seed round Monday, led by Touring Capital with participation from Principal VC and individual investors from OpenAI and Anthropic. The San Francisco company is building CUDA-alternative kernel software that runs on any chip architecture — GPUs, SRAMs, phone chips, and systolic arrays — aiming to break the lock-in that ties AI workloads to Nvidia's proprietary compute stack. Infinity was founded by engineers with backgrounds in low-level systems and compiler design. The round values the company at $100 million pre-money. Demand for chip-agnostic inference software has grown as enterprises seek to diversify beyond Nvidia amid persistent GPU supply constraints and long lead times.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Tech Pulse
Top Frontier Models: Claude Mythos 5 (95.5% SWE-bench) | Claude Fable 5 (95.0% SWE-bench) | GPT-5.6 Sol (~94.6% GPQA)
Top Open Source Models: GLM-5.2/Z.ai (91.2% GPQA Diamond) | Kimi K3/Moonshot (81.2% FrontierSWE) | DeepSeek V4 (80.6% SWE-bench Pro)
Top Small Models (15–50B): Qwen3.5-32B | Phi-4.1-32B | Llama 4 Maverick
Top Edge Models (0–15B): Phi-4/14B (84.8% MMLU) | Gemma 4 E4B (83.7% ARC) | Qwen3.5-9B
AI Leaders: NVIDIA $4.91T | Alphabet $4.31T | Microsoft $2.83T
Robotics Leaders: Intuitive Surgical ~$175B | ABB ~$165B | Fanuc ~$50B
Curated by JD · samwise.agency