GitHub Actions is changing the infrastructure that runs CI jobs while tightening the requirements for teams operating their own runners. In September and October 2026, GitHub introduced more detailed runner-version deprecation information, began enforcing updated self-hosted runner requirements for Enterprise Cloud, and expanded eligible hosted Linux jobs across a multi-cloud compute pool. The announcements affect different groups of customers, but together they make runner infrastructure an increasingly explicit part of delivery reliability planning.
A workflow can keep the same YAML and still encounter a new operational dependency. Hosted runners may use different outbound IP addresses as their underlying cloud changes. Self-hosted runners can stop receiving jobs when their installed software falls behind the required version. Neither situation is necessarily caused by an application change, and both deserve attention from the engineers responsible for network access, runner images and deployment continuity.
Hosted Linux runners can now use more than one cloud
On October 28, GitHub announced that eligible standard 2-core and 4-core Linux runners, along with compatible larger runners, can be scheduled on AWS as well as Azure. GitHub says the expanded pool is intended to improve available capacity, queue times and resilience. Existing labels and workflow syntax remain unchanged. For Free, Pro and Team customers, eligible jobs are automatically included; Enterprise Cloud customers must opt in through enterprise Actions policies. Windows, private-networked and other specialized runners are excluded from this particular change.
The additional compute capacity is useful, especially for teams with variable CI demand, but the benefit is a vendor objective rather than a guaranteed improvement for every repository. Queue time depends on job mix, concurrency limits, runner availability and other factors. Teams should compare their own queue and execution measurements before concluding that the change has improved delivery performance.
Network policy is a more immediate operational consideration. GitHub explicitly warns that the expanded pool introduces outbound IP addresses beyond Azure and Microsoft-owned ranges. A job may start normally and then fail when it tries to reach a registry, database, private service or identity provider that permits only previously recognized source addresses. Microsoft Entra Conditional Access policies can be affected by the same assumption. GitHub recommends using the `actions` array from its public `GET /meta` API to maintain allowlists rather than filtering only by cloud-provider ownership.
Allowlisting every published Actions address may be inappropriate for sensitive destinations because those addresses are shared infrastructure and can change. For systems requiring stronger network isolation, teams should assess dedicated networking, private endpoints, short-lived identities and restricted runner groups. The correct approach depends on which services the workflow must reach and how much trust the destination places in source IP alone.
Self-hosted runner versions now have an operational deadline
GitHub's September 28 update moved the Enterprise Cloud enforcement date to September 29, 2026. The registration requirement is version 2.329.0 or newer, while the version needed to keep executing jobs is a separate, moving requirement. GitHub's published policy requires self-hosted runners to receive new releases within 30 days; a critical security update can require faster action. GitHub Enterprise Server is outside the scope of this particular enforcement announcement.
This distinction is especially relevant for organizations using immutable VM images, custom container images or Actions Runner Controller. A runner may have registered successfully months earlier and still become unable to receive new work because its runtime is no longer supported. Autoscaling will not repair the problem if every newly created pod or VM starts from the same outdated image. The issue may first appear as queued jobs, even though Kubernetes nodes, runner scale sets and the underlying cloud infrastructure seem healthy.
Runner upgrades therefore belong in normal platform maintenance. Teams should inventory runner software versions, identify which images and templates supply them, and ensure the release process can update those artifacts promptly. They should also test runner changes against commonly used actions, container tooling, authentication methods and build dependencies before broad rollout. Where auto-update is disabled, a scheduled rebuild and rollout process is necessary to keep the fleet compatible.
Deprecation data can support earlier detection
GitHub's September 3 Actions update added a REST endpoint for runner-version deprecations at repository, organization and enterprise scope. The response includes the runner version and separate dates for registration and runtime deprecation. This gives platform teams a way to detect approaching incompatibility before workflows begin waiting for unavailable capacity.
An effective check would compare installed versions across runner groups with the published deprecation information, then alert on fleets nearing the runtime deadline. It should also account for image-build lead time, staged rollout windows and the ability to recover if an upgrade breaks a common workload. Registration audit events are useful for identifying newly registered versions, but they do not provide a complete inventory of all connected runners. Runtime inventory still needs a reliable source of truth.
The same September release introduced narrower `GITHUB_TOKEN` permissions for reading vulnerability alerts and source-identity fields for reusable workflows. These are separate features, yet they reinforce the value of explicit execution context: platform operators can reduce unnecessary privileges and distinguish the workflow being executed from the repository event that invoked it. Such controls are particularly relevant when reusable workflows and centrally managed runner groups are shared across many teams.
What platform teams should review
For hosted jobs, we would start with external dependencies: registries, package mirrors, cloud authentication, internal APIs and any firewall or identity policy that assumes a fixed runner network. Confirm which jobs are eligible for multi-cloud placement, check current GitHub Actions address ranges, and run representative jobs through the same authentication and deployment paths used in production. Track queue duration separately from job execution time so scheduling improvements and application-level failures remain distinguishable.
For self-hosted fleets, inventory runner versions and image provenance, then test whether new capacity can register and pick up work. ARC deployments deserve a check of runner container images and upgrade automation, including any settings that disable automatic updates. Alerts should identify the affected runner group and the corresponding repositories rather than reporting a generic CI failure. Teams can also maintain a small tested rollback image, provided that image remains above GitHub's supported minimum.
These changes affect engineering throughput because a build pipeline depends on more than source code and test scripts. A runner's software version, network identity and scheduling environment can determine whether an otherwise valid workflow reaches its dependencies. We would treat those properties as reviewable infrastructure configuration, with owners, change history and measurable failure modes.
At InfraJump, infrastructure inventory, resource inspection, security findings and cost context can help teams understand the systems that support delivery workflows. The runner-specific checks still need to be implemented and verified within each organization's GitHub and cloud environment. For teams operating a mix of hosted and self-hosted capacity, the recent Actions changes are a useful reason to review those boundaries before the next routine upgrade or network-policy change.
Primary sources
- GitHub, September 3, 2026 — [Early September Actions updates](https://github.blog/changelog/2026-09-03-github-actions-early-september-2026-updates/) - GitHub, September 28, 2026 — [Self-hosted runner enforcement date moved](https://github.blog/changelog/2026-09-28-self-hosted-runner-version-enforcement-date-has-moved/) - GitHub, October 28, 2026 — [Multi-cloud compute for Actions](https://github.blog/changelog/2026-10-28-github-actions-adds-multi-cloud-compute-to-improve-reliability/) - GitHub, June 12, 2026 — [Runner version enforcement background](https://github.blog/changelog/2026-06-12-github-actions-minimum-version-enforcement-timeline-for-self-hosted-runners/)


Comments (0)
Add a comment