エンタープライズチーム向けモデルポリシーターゲティングがパブリックプレビューにEnterprise teams model policy targeting in public preview

GitHub has introduced user-based model policy targeting in public preview, giving AI administrators a more granular way to control which underlying models their developers can access through Copilot. The capability is available to GitHub Enterprise customers that hold Copilot Business or Copilot Enterprise licenses, and it matters because it moves model governance away from a single organization-wide switch toward policies that can differ from one team to the next.

Until now, administrators who wanted to permit or restrict specific AI models typically applied one rule across an entire enterprise or organization. That approach is simple but blunt: a policy strict enough for a sensitive group may be too restrictive for others, while a permissive default may expose teams that need tighter controls. The new targeting feature is designed to resolve that tension. According to the changelog, it empowers AI administrators to set a baseline policy for the organization and then grant particular teams different model access on top of it.

In practice, this means an administrator can define a default set of models that everyone may use, then selectively broaden or narrow that set for individual teams. A research or platform team might be allowed to experiment with newer or more capable models, while a group handling regulated data could be limited to an approved subset. Because the policies are layered, the baseline continues to apply wherever a more specific rule has not been assigned, which reduces the risk of teams falling outside governance entirely. This layered model is a common pattern in enterprise administration, and applying it to AI model selection reflects how organizations increasingly treat model choice as a governance decision rather than a purely technical one.

The feature sits within Copilot's broader administrative controls, which already let enterprises manage policies such as whether Copilot is enabled, how suggestions are handled, and which features are available to members. Model policies specifically govern access to the range of models that Copilot can route requests to. Over the past year, GitHub has expanded Copilot to support multiple model providers, so an enterprise may have access to several families of models rather than a single default. As that catalog grows, the ability to decide who can use which model becomes more consequential, and per-team targeting appears intended to give administrators the precision to match model access to team needs and risk profiles.

It is worth noting the public preview status. Preview features are generally made available so that customers can test functionality and provide feedback before general availability, and they may change based on that input. Organizations evaluating the capability should treat it as still evolving, with behavior, interface details, and scope potentially subject to revision before a stable release. GitHub has not, in this announcement, indicated pricing changes tied to the feature; it is offered to existing Copilot Business and Copilot Enterprise customers as part of their licensing.

For administrators, a few prerequisites are relevant. Model policy targeting depends on having teams defined within the enterprise or organization structure, since policies are assigned to those teams. It also requires the AI administrator role or equivalent permissions to configure policies, which keeps control in the hands of designated governance owners rather than individual developers. Enterprises that have already invested in organizing members into teams for repository access and other permissions are likely to find the targeting model familiar, because it extends an existing structure rather than introducing a separate one.

The move fits a wider industry trend in which vendors of AI coding assistants are adding enterprise governance controls alongside raw capability. As competing tools and platforms expand their model options, buyers are increasingly asking not just what a tool can do but how access can be scoped, audited, and aligned with internal policy. Team-level model targeting is a response to that demand, letting a single enterprise accommodate different regulatory, security, and productivity requirements without fragmenting into separate deployments. For teams weighing how to roll out Copilot at scale, the feature offers a way to standardize a safe default while still allowing flexibility where it is warranted, though its ultimate value will depend on how it performs as it moves beyond preview.