AI & Tech Brief — 22 September 2026: Model Choice, Molecular Synthesis and AI Adoption

Edition date: 22 September 2026 (JST). This is a retrospective edition published on 23 September 2026 (JST) at the publisher's request. The source announcements below were published on 21 September; they are not presented as 22 September releases.

Overview

Three September 21 updates show different ways AI is moving from model availability to practical evidence: a new coding-model option in GitHub Copilot, a research system that proposes chemical synthesis routes, and a measured view of global AI adoption. Each claim has a different maturity level, from a gradual product rollout to benchmark research and telemetry-based estimates.

Key takeaways

• GitHub began a gradual rollout of Grok 4.7 in Copilot, with administrator controls for business plans.
• Microsoft researchers reported promising benchmark results for RetroChimera, while real-world discovery evaluation remains ahead.
• Microsoft estimated generative-AI use by 18.8% of the global working-age population in June 2026, with a continuing North–South gap.

Top stories

1. GitHub adds Grok 4.7 to Copilot's model options

Source published 21 September 2026. GitHub says Grok 4.7 is rolling out for Copilot Pro, Pro+, Max, Business, and Enterprise. Its model picker will expose the model across supported editors, the Copilot CLI, cloud agent, and app. Rollout is gradual; an eligible user may not see it immediately. GitHub says usage-based billing follows provider list pricing, and Business and Enterprise administrators can manage access through model policy.

Primary source: GitHub Changelog — Grok 4.7 in GitHub Copilot

2. RetroChimera explores AI-assisted synthesis planning

Source published 21 September 2026. Microsoft Research and collaborators describe RetroChimera, which combines complementary model predictions with search to propose routes from a target molecule back to available building blocks. In one 10-molecule benchmark, expert chemists fully accepted its proposed sequence for nine molecules, versus two to five for comparator systems. That is a research result, not proof of routine laboratory or clinical performance; the team says evaluation in real discovery settings is next.

Primary source: Microsoft Source — RetroChimera research

3. Microsoft's AI-diffusion estimate exposes an adoption gap

Source published 21 September 2026. Microsoft's Q2 2026 report estimates that 18.8% of working-age people worldwide used a generative-AI product in June, about one percentage point above Q1. Its estimated rates were 28.8% in the Global North and 16.2% in the Global South. The measure comes from aggregated, anonymized Microsoft telemetry adjusted for market share, connectivity, and population; it is an estimate rather than a census of all AI use.

Primary source: Microsoft On the Issues — Global AI diffusion in 2026

Watchlist

• How quickly Grok 4.7 reaches eligible Copilot users, and how organizations set model access policies and spending controls.
• Whether RetroChimera's benchmark performance carries into chemists' real discovery workflows.
• How broader coverage of AI tools changes future adoption estimates and regional comparisons.

Primary sources checked on 23 September 2026 (JST). The source publication dates are stated above; this 22 September edition was published retrospectively on 23 September.