On July 31, 2026, GitHub announced user-based model policy targeting for Copilot Business and Enterprise customers in public preview. An enterprise can set a baseline and then make additional models available to selected enterprise teams, giving different roles room for newer models or frontier experimentation.
The policy has three enterprise-level states: enabled, disabled, and optional. An optional model can then be assigned to particular enterprise teams. GitHub says most enterprise customers can opt into the rollout on August 3. The structure changes model adoption from a single company-wide switch into a policy that can be aligned with roles, training, and bounded experiments.
Team targeting is not complete permission isolation. GitHub’s rule is least-restrictive: if a user receives a model through any one enterprise team, that model is available to the user elsewhere. Once enterprise teams mode is enabled, organization-level model settings no longer apply, although rollback is available during the preview. The model policy applies only to users whose license is assigned by that enterprise.
For administrators, the value is a single place to connect model trials, capability baselines, and cost governance. A team can test a newer model before a broader rollout, but team membership, enterprise roles, license ownership, and data policies still need separate checks. A team assignment should not be treated as proof of data isolation or compliance.
GitHub’s update shows enterprise AI governance moving from allowing or blocking a model toward more granular policy layers. Before rollout, organizations should record each model’s purpose, cost, data-handling requirements, approver, and rollback path. They should also test which policy wins when a user belongs to multiple enterprise teams.
