Samwise Tech/AI/Robotics Newsletter
Sunday, August 30, 2026
Sony Music, Warner Chappell sue Anthropic over alleged ‘brazen campaign’ of copyright theft
Sony Music Publishing and Warner Chappell Music have filed suit in U.S. District Court for the Northern District of California against Anthropic and co-founders Dario Amodei and Benjamin Mann, alleging a ‘brazen campaign of illegally torrenting, scraping, and downloading copyrighted works.’ The complaint accuses Anthropic of ‘flagrant piracy,’ including using illegally torrented books containing lyrics and sheet music to train its AI models. The lawsuit follows a prior Bartz case that settled for $1.5 billion, signaling that music publishers are turning to litigation as their primary strategy for extracting licensing agreements from AI developers.
Sources: TechCrunch
Anthropic wins first court ruling against Pentagon’s supply-chain risk designation
U.S. District Judge Rita Lin has ruled that the Trump administration’s designation of Anthropic as a supply-chain risk was illegal, finding the label constituted ‘unlawful retaliation’ against the company, was ‘arbitrary and capricious,’ and violated Fifth Amendment due process. The ruling stems from Anthropic’s refusal to remove safety guardrails for autonomous weapons and mass surveillance applications — refusals the administration allegedly used as grounds for the designation. Judge Lin also noted a significant contradiction: the Department of Defense simultaneously continued pursuing a contract with Anthropic and collaborating on the Mythos AI model. A related D.C. lawsuit remains ongoing.
Sources: TechCrunch
Nvidia’s AI advantage is moving beyond the GPU
Nvidia is rolling out the Vera Rubin architecture, pairing its Rubin GPU with the Vera CPU, a Groq 3 LPX inference accelerator, and dedicated storage and networking racks. The Vera CPU handles data orchestration for megascale data centers, where getting stored data to the GPU efficiently has become a central challenge. ‘We saw upwards of 3x improvement in these operations, where the Vera CPU is allowing for acceleration,’ Jason Hardy, Nvidia’s VP of storage technology, told TechCrunch. Meanwhile, OpenAI’s Jalapeño chip takes the opposite approach: eliminating data movement entirely within one integrated system. Both strategies reflect AI’s shift from raw compute to system-level efficiency.
Sources: TechCrunch
Open-weight AI companies are Silicon Valley’s hottest acquisition targets
Open-weight AI companies have become Silicon Valley’s most sought-after acquisition targets. Nvidia is reportedly closing in on a $13 billion acquisition of Hugging Face and has already struck a $6 billion deal with Poolside. Stripe acquired OpenRouter for over $7 billion. Despite these valuations, adoption remains narrow: only 6% of companies use open-weight models, according to a Ramp survey, with just 2% of software engineers running them, per Jellyfish. The driver is repeated inference workloads — customer service, internal tooling — where open weights beat proprietary costs. ‘Tokens are the central currency for companies building with AI,’ Stripe’s Patrick Collison said. Fireworks AI processes 40 trillion tokens per day.
Sources: TechCrunch
Chinese automakers are following Tesla’s bet that robots are the next big profit machine
Chinese automakers are racing into humanoid robotics, following Tesla’s lead in treating robots as a future profit engine. Xpeng’s robotics unit raised over $900 million at a valuation exceeding $6.3 billion, backed by IDG Capital, Tencent, and Alibaba, while the founder and co-president invested $100 million personally. BYD unveiled its humanoid robot, Xiao Di. AiMOGA, backed by Chery Automobile, is preparing an IPO. Changan, GAC, Li Auto, SAIC, and Seres are also developing robots. Outside China, Hyundai plans to deploy Boston Dynamics’ Atlas at its Georgia factory, and Mobileye acquired Mentee Robotics for $900 million.
Sources: TechCrunch
Meta researchers taught an 8B model to match Claude Opus 4.5 — without the frontier price tag
Researchers from Meta AI and the University of Illinois have introduced EvoHarness-RL, a framework that trains smaller AI models to perform at frontier-model levels on agentic tasks. At the core is a BPE (Belief, Progress, Experience) interface giving agents four meta-actions — track, commit, recall, and note — to build working memory during task execution. Qwen3-8B trained with the system achieved 96.9% on ALFWorld, a 49-point improvement that matches Claude Opus 4.5’s 96.4%. GPT-4.1 improved 22.1 points and GPT-5 improved 25.7 points using the BPE framework. The team also developed ‘harness annealing,’ reducing tool use as the agent masters routines.
Sources: VentureBeat
An Anthropic researcher previews self-improving AI that outperforms human alignment experts
An Anthropic fellow, Chen Yueh-Han, has led a paper introducing the Automated Alignment Researcher, or AAR — a system designed to improve AI models’ alignment properties without human intervention. The paper, ‘Automated Researchers Can Reliably Mitigate Alignment Failures,’ demonstrates that AAR improved performance on 10 alignment benchmarks without degrading overall model capability. The system works iteratively: searching research literature, proposing new methods, and training the model — each iteration taking roughly 30 minutes. The best AAR method outperforms solutions proposed by experienced human researchers ‘on average within six hours’ while costing $4 per hour, compared to $150 per hour for human researchers.
Sources: TechCrunch
Every newsletter preserved and searchable
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.