Two GitHub announcements published on 27 August 2026 affect opposite ends of the software delivery process: how code receives automated scrutiny, and how long the records of its checks survive. Our assessment is that teams should address them together. More review coverage is useful, but it does not replace accountable approval or a deliberate record of why a release was accepted.
What happened: broader Copilot review
GitHub says Copilot code review now supports automatically requested reviews of bot-authored pull requests when the relevant organisation policy is enabled, with usage billed to the organisation. Pull requests opened by Copilot cloud agent can receive full agentic review. The previous 300-file or 20,000-line size limit has also been removed.
Reviewers can now classify resolved Copilot comments as Addressed, Won't fix or Incorrect. GitHub describes these reasons as feedback to improve the product. They are not a guarantee that a fix is correct, nor a substitute for the responsible team's release decision.
What happened: an evidence-retention deadline
From 1 October 2026, GitHub says checks, workflow runs and statuses will follow the Actions artifact-and-log retention setting, whose default is 90 days. Previously, those records remained for more than 400 days regardless of that setting. Public repositories have a maximum of 90 days; organisation and enterprise caps also constrain repository settings.
Raising retention will not restore records already removed. Checks, workflow-run and status metadata are not billed for storage, but extending the shared setting also keeps associated artifacts and logs longer and may increase storage charges. GitHub advises exporting records that must outlive the configured window.
Why it matters
Our interpretation is that automated review and evidence preservation solve different problems. Review helps identify issues before merging. Retained evidence helps explain a decision later, when an incident, customer question or maintenance task requires reconstructing what was tested. A green result that can no longer be retrieved is a weak foundation for that reconstruction.
The expanded size support should not encourage larger, harder-to-understand changes. Keep pull requests focused enough for a person to judge intent and risk. Treat an AI author and an AI reviewer as useful tools, not independent human accountability; shared blind spots still need tests, domain expertise and clear approval rules.
The practical implication
Before 1 October, inventory the repositories that support customer-facing or otherwise consequential systems. Assign an owner to decide which release records must remain available and for how long. Where the platform window is insufficient, design an access-controlled archive containing the commit identifier, relevant test outputs, approval decision and artifact identifiers. Test that the archive can actually answer a sample release-history question.
Separately, review the policy and budget for automatic bot-PR reviews. Use resolution reasons consistently, and add a short human explanation when dismissing a material finding. Measure whether reviews surface useful defects and whether teams can reconstruct releases—not simply how many comments the automation produces.
Key takeaways
Set evidence-retention requirements before the 1 October 2026 change takes effect.
Keep human release ownership even when both code generation and review use AI.
Check review billing and archive retrieval, not just whether automation is enabled.
Sources
This article is original analysis written by IT Web Builder. The reporting and product details below were used as references and are credited to their publishers.
GitHub Changelog — Copilot code reviewResolution reasons and expanded capabilities — published 27 August 2026↗GitHub Changelog — Actions retentionActions retention will cover checks, workflow runs, and statuses — published 27 August 2026; effective 1 October 2026↗IT Web Builder publishes original summaries and practical commentary. We do not reproduce source articles. Product claims and announcements are linked to their original publishers so readers can verify the details.