Newsletter Archive

Anthropic rolled out autonomous computer control for Claude's Code and Cowork tools, letting the AI open files, browse the web, run dev tools, and execute tasks without manual prompting — available now as a research preview for Pro/Max subscribers on macOS.

The Verge AI · 3 min read Industry

Mistral AI launches Forge, a system letting enterprises train frontier-grade models on proprietary internal data instead of generic public datasets. Already partnered with ASML, ESA, Ericsson, and others to embed domain knowledge—engineering standards, compliance policies, codebases—directly into custom models.

Mistral AI · 6 min read Industry

Heather Kulik, a pioneer in AI for materials discovery, explains why there's no single 'AlphaFold moment' for materials science and shares hard-won lessons from a decade combining computational tools with domain expertise. Her group recently used AI to design polymers with novel quantum mechanical properties that surprised the experimentalists who built them.

Latent Space · 6 min read Community

Anthropic rolled out computer use for Claude, letting the model control PCs to complete tasks like clicking, typing, and navigation. This moves beyond text-based AI toward genuine task automation that developers can build into applications.

PYMNTS · 3 min read Industry

The White House released a comprehensive national AI policy framework establishing governance priorities, safety standards, and regulatory guidelines that will influence how companies develop and deploy AI systems across the U.S.

Jdsupra · 10 min read Policy

Arm is manufacturing its first proprietary CPU, the Arm AGI CPU, for AI inference workloads, with Meta as lead partner and first customer. This marks a major strategic shift: after decades licensing designs only, Arm is now competing directly in the datacenter chip market alongside Nvidia and AMD, and Meta gains leverage to reduce AI infrastructure costs through diversification.

The Verge · 3 min read Hardware

A federal judge called the Pentagon's campaign to designate Anthropic as a supply chain risk "troubling" and suggested the three-pronged Trump administration action (presidential ban, contractor restrictions, and supply chain designation) appears designed to "cripple" the company rather than address legitimate security concerns. Anthropic is fighting to pause the designation and restore pre-ban status while litigation proceeds.

Axios · 2 min read Policy

Claude Code now has auto mode, where Claude makes permission decisions autonomously with a separate Sonnet 4.6 classifier monitoring actions before execution. The safeguard system blocks actions that escalate beyond task scope, target unrecognized infrastructure, or appear triggered by hostile content in files—and developers can customize filters with their own rules.

Simon Willison · 4 min read Tools