AI roundupFriday 28 August 2026

Judge blocks Pentagon's Anthropic blacklist

A federal judge ruled the Pentagon's attempt to blacklist Anthropic was unconstitutional retaliation, marking a significant legal win for the AI company. Meanwhile, the rest of the day was about infrastructure: a new benchmark for science AI agents, a playbook for AI-native software development, and Brazil threading a needle between US and Chinese tech.

In today's edition: Anthropic wins in court, Benchmarks and playbooks and Global compute chess, plus quick bites.

Anthropic wins in court

The biggest story today is a legal check on executive power over AI companies.

Benchmarks and playbooks

Two concrete efforts to define how AI should work in the real world appeared today.

terminal-bench-science.ai · 17 min

0
TERMINAL-BENCH-SCIENCE

Stanford and domain experts released Terminal-Bench-Science, a benchmark that tests AI agents on real scientific workflows rather than synthetic tasks. It puts scientists in charge of defining what capability looks like.

claude.com · 39 min

0
The AI-Native SDLC playbook

Anthropic published a playbook for adapting the software development lifecycle to AI-native tools like Claude Code, arguing that traditional approval gates are now the bottleneck, not code generation.

Global compute chess

A few pieces today show the geopolitics of AI infrastructure playing out at different scales.

Quick bites