Anthropic blocks advanced model access after White House clash

Key Shifts

Anthropic blocks advanced model access after White House clash: Commerce Secretary Howard Lutnick sent a letter demanding Anthropic restrict foreign access to frontier models like Claude Fable 5. Anthropic disabled its Mythos model in response. The Trump administration blocked G7 allies from accessing Anthropic’s most advanced models, and the Pentagon is reducing its reliance on Anthropic in favor of competitors. AI governance is rapidly evolving into a trade and national security landscape. Wired · NY Post · Crypto Briefing · Bloomberg · Techdirt
SpaceX acquires AI coding tool Cursor for ~$60B: Elon Musk’s SpaceX has purchased Cursor, a competitor to Claude Code and OpenAI Codex. The AI coding tools market is being elevated from developer productivity to strategic national-security and space-infrastructure asset territory. 9to5Mac · Pentesty

Anthropic “pauses” token-based billing for Claude Agent SDK: Agent usage grew far faster than expected, forcing Anthropic to halt token-based billing while it rethinks the pricing model. A clear signal that agent economics remain unsettled. Ars Technica
Z.ai releases GLM-5.2, an open-weights frontier model: Chinese lab Z.ai published GLM-5.2 on HuggingFace with open weights. The open-source AI camp is closing the frontier-performance gap rapidly, and the global model supply chain is fragmenting into multiple poles. HuggingFace
Netherlands unveils GPT-NL, a sovereign language model: Dutch research institute TNO launched GPT-NL, a domestically built LLM. The EU’s AI sovereignty strategy is moving from policy papers to deployed products. TNO
GitHub Models discontinued for new customers: GitHub is shrinking its AI model marketplace, cutting off new customer sign-ups. Big tech is beginning to prune AI product portfolios. GitHub Blog
Cursor Origin: a Git forge built for the agentic era: Cursor unveiled Origin, a version-control paradigm that integrates code review, testing, and deployment into agent-native pipelines — the first VCS designed for workflows where AI agents are the norm. Cursor

Agentic coding amplifies expert advantage, not beginner gains — Anthropic research: Analysis of ~400,000 Claude Code sessions shows agentic coding tools deliver larger productivity boosts to experienced developers than to novices. Counter-narrative to “AI replaces expertise.” The gap between skilled and unskilled practitioners may widen, not shrink. Anthropic
Qwen-Robot Suite: a foundation model for physical-world intelligence: Alibaba’s Qwen team released a foundation model suite for robot control, unifying vision, language, and action across diverse robot form factors. The robotics foundation-model race is accelerating. Qwen
AI Index Report 2026 published: Stanford HAI’s annual benchmark report. This year’s edition highlights AI safety governance metrics and open-source ecosystem trends. arXiv

Running local models is genuinely good now — time to experiment: Vicki Boykis makes the case that local LLM inference has reached practical usability in 2026, including on Apple Silicon. Suitable for privacy-sensitive enterprise use cases and air-gapped environments. Now is the right time to start AI PoCs that don’t touch the cloud. Vicki Boykis
Offloading AI models to SSD to bypass the RAM bottleneck: Lexar demonstrated technology to run local AI models by leveraging SSD storage instead of scarce RAM, opening a path to reduce dependence on expensive GPUs and HBM. TechPowerUp

German court holds Google liable for AI-generated falsehoods: A German court ruled that Google bears legal responsibility for false information in its AI-generated answers. Legal liability for AI content is beginning to diverge by jurisdiction — startups operating globally must track this fragmentation. DW
Microsoft sued by shareholders over AI and cloud spending: Shareholders are challenging Microsoft’s AI investment returns, marking the first market pressure test on whether big tech’s AI capex rally is sustainable. Reuters