GitHub Issues Adds Approval, Confidence, and Rationale Controls for Agents
GitHub Issues now grades agent changes by confidence and lets people approve them, but GitHub says this is not a security control.
What shipped
GitHub has released three linked capabilities inside GitHub Issues, now in public preview: Approvals, Confidence, and Rationale. Together they let an automation label, type, assign, or close issues while showing a person the reasoning behind each change before or after it applies. According to GitHub's changelog, the goal is to let a team decide how much to automate and when to stay in the loop.
How the three pieces work
Approvals let you tell an automation to suggest a change instead of applying it directly; the suggested change waits in a panel on the issue until someone accepts or declines it, one at a time or all at once. Confidence has agents rate every supported action as high, medium, or low; high-confidence changes apply automatically, while GitHub Issues holds medium- and low-confidence ones as suggestions for review. Rationale records the reason behind every supported action, whether it ran automatically or waited for a decision, giving an audit trail of what changed and why.
The caveat GitHub wants readers to notice
GitHub is explicit that approvals are a workflow convenience, not a security control. The company states plainly that approvals do not enforce a server-side boundary, and an agent that already has permission to change issues can apply changes directly instead of only suggesting them. In other words, the review panel shapes normal automation behavior, but it is not a permissions wall that stops an agent from acting.
Where this works today
Rationale, confidence, and approvals work with GitHub Agentic Workflows and Copilot cloud agent automations, and the same data is available through the REST and GraphQL APIs. At launch, the supported changes cover labels, fields, type, close, and assignees. Teams already using GitHub Agentic Workflows can upgrade their workflows to add issue intent support; supported safe outputs then include intent information automatically while staying backward compatible.
Key facts
- Three new capabilities: Approvals, Confidence, and Rationale, all in public preview.
- Confidence levels are high, medium, or low; only high-confidence changes apply without review.
- Repository admins can configure the automation level to set the confidence threshold for what applies automatically.
- GitHub states approvals are a workflow convenience, not a security control, and do not enforce a server-side boundary.
- Supported safe outputs for issue intents: set-issue-type, set-issue-field, add-labels, close-issue, assign-to-agent, and assign-to-user.
What you can check today
Use has:suggestions in GitHub Issues search to find issues with changes waiting for your review. If you already run GitHub Agentic Workflows, add issue-intents: true to a workflow's frontmatter to require intents for a specific safe output, or set it to false to opt out. To try the feature from scratch, create an automation from the Automations pane in your repository's Agents tab; no update is required first. Because approvals are described as a convenience rather than a boundary, treat repository permissions, not the approval panel, as your actual access control.
Check your understanding
- What are the three new capabilities GitHub added to GitHub Issues, and what does each one do?
- Why does GitHub say approvals should not be treated as a security control?
- Which setting can a repository admin adjust to change which confidence level of actions applies automatically?