Signal, not noise: five primary-source reads from September 18, each with a two-line note on why it matters for builders.
Copilot code review ships an improved review experience
Copilot code review now groups findings into open, resolved-since-last-review, and previously-missed buckets with severity labels and deep links, auto-resolves its own comments more carefully, and generates commit messages when you land a batch of suggestions. Turn the overview comment into your review triage queue so human attention lands on genuinely open findings first.
Read the full story at GitHub Changelog (September 18, 2026).
Code coverage ruleset conditions are now API-manageable
The restrict-code-coverage ruleset option, previously UI-only, is now generally available through the REST API, so you can enforce minimum line coverage or cap coverage drops programmatically. Codify the rule in your infrastructure-as-code setup to keep coverage gates consistent across every repository instead of drifting per-repo.
Read the full story at GitHub Changelog (September 18, 2026).
Stage-only npm tokens split staging from publishing
New granular npm tokens with read-and-write stage-only scope let automation run npm stage publish while direct npm publish stays rejected, with a maintainer approving release under 2FA. Adopt stage-only tokens now as the migration path before direct publishing through bypass-2FA tokens is removed, targeted for January 2027.
Read the full story at GitHub Changelog (September 18, 2026).
Selected Copilot models retire on October 19
GitHub will deprecate Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 across all Copilot experiences on October 19, 2026, with named successors such as Gemini 3.8 Flash and GPT-5.6 variants. Pin your workflows and model policies to the suggested alternatives now so completions and agents do not silently shift under you.
Read the full story at GitHub Changelog (September 18, 2026).
Anthropic pilots embedded evaluation with Accenture
Anthropic is partnering with Accenture, led by its Faculty unit, to embed independent evaluators inside the lab with employee-comparable access to red-team models, assess alignment, and verify safety commitments, with each side investing at least $1 billion over five years. If you ship safety-critical AI features, watch this model: continuous embedded oversight may soon be the bar your enterprise customers expect.
Read the full story at Anthropic (September 18, 2026).