Ask Cat › AI Tool Summary › GitHub Copilot
Copilot's Global Model Policy Goes Live Before Sept 1: Models You Never Configured Get Switched On — and the 4 Kinds That Don't
Article last updated:2026-08-27
If you’re the person who administers GitHub Copilot at your company, this is a good week to open the model list and look at it.
On 2026-08-26 GitHub’s official changelog announced that the global model policy previewed in July is now being enforced, rolling out in waves, with the rollout completing by September 1 (GitHub Changelog, 2026-08-26).
In one sentence: every model you never explicitly configured will start following a default value — and that default is “enabled.”
Here is what the official sources actually say, broken into four questions: who it affects, what gets switched on, what never does, and what admins should do now.
1. Who this affects: Business and Enterprise only
The changelog is specific about scope: the policy covers “generally available GitHub Copilot models on Copilot Business and Copilot Enterprise plans.”
The official reference page, About default availability of Copilot models, lists the same two plans under “Who can use this feature?”
So:
| Your plan | This policy |
|---|---|
| Copilot Free | Not on the official list of affected plans |
| Copilot Pro (US$10/mo) | Not on the official list of affected plans |
| Copilot Pro+ (US$39/mo) | Not on the official list of affected plans |
| Copilot Max (US$100/mo) | Not on the official list of affected plans |
| Copilot Business (US$19/seat/mo) | Affected |
| Copilot Enterprise (US$39/seat/mo) | Affected |
(Plan prices are from this site’s verified GitHub Copilot tool page. Since 2026-06-01 Copilot bills as a subscription plus GitHub AI Credits usage.)
Individual users aren’t free of model restrictions — theirs just come from a different mechanism. This site’s verification log records that since 2026-06-24, Free and Student plans can only use “Auto” model selection and cannot pick a model by hand; Pro and above are unaffected. That’s a separate rule, not this policy.
2. What gets switched on
Three things happen once the policy takes effect for your org (changelog):
- Models you never configured flip to “Delegate to default policy” and begin following your policy setting. If the policy is enabled — and enabled is the default — those models become available to your users.
- “Delegate to default policy” is a live state, not a one-time migration. GitHub’s wording: a “live, dynamic state that always tracks your policy.” Change the policy later and every model in that state changes with it.
- Explicit choices are preserved. GitHub commits to “We always preserve explicit choices” — anything you deliberately enabled or disabled stays exactly as you set it.
There’s a consequence that’s easy to miss: the official docs state that every future model release inherits this default until you explicitly configure it. This isn’t just a one-off batch change; it’s the standing behaviour for new models from here on.
After the rollout, each model shows one of four states:
| State | Meaning |
|---|---|
| Enabled | You explicitly turned it on |
| Disabled | You explicitly turned it off |
| Delegate to enterprise teams/apps or organizations | Follows a setting inherited from your enterprise team or org |
| Delegate to default policy | Follows your default enablement policy (the new one) |
3. The 4 kinds of models that are never auto-enabled
This is the part worth writing down. About default availability of Copilot models lists four categories that are disabled by default regardless of how your policy is set:
- Pre-GA models
- Open weight models — GitHub names DeepSeek, Kimi K2.7 Code, and Kimi K3 as examples
- Models not covered by GitHub’s data retention agreement — GitHub names Claude Fable 5 as the example
- For enterprises restricted to data-resident or FedRAMP-compliant models, any model that doesn’t respect those policies
The changelog gives a shorter version of this list (only categories 2 and 3). The docs page is the more complete list — admins should work from that one.
Note what the third criterion actually measures: not whether a model is good, but whether it falls under GitHub’s data retention agreement. It’s a compliance boundary, not a quality ranking — which is why a widely used commercial model sits on the same exclusion list as open-weight ones.
4. Three things admins should do now
Per Managing availability of models in your enterprise:
Step 1: Look at the current state. Go to your enterprise → click AI controls at the top of the page → click Copilot in the sidebar → click Configure models. Anything labelled “Delegate to Default Policy” is what this rollout will touch.
Step 2: Make the important decisions explicit. GitHub’s own recommendation is to explicitly configure every model that should be on for everyone or off for everyone, establishing a baseline; leave delegation for the models where only some people should have access. Explicitly configured models are immune to this change and to future policy changes.
Step 3: If you want “off by default,” turn the policy off. Disable the Default availability for released models policy in your enterprise or organization model policies, and unconfigured models default to disabled. GitHub notes you can disable it in just the organizations with stricter compliance requirements rather than enterprise-wide.
5. GitHub has already flagged that this state may go away
The changelog’s “What’s next” section deserves its own note: GitHub says it is evaluating making the global model policy an explicit decision and removing the “Delegate to default policy” state altogether, so that every policy reflects “an explicit, intentional choice rather than an inferred one.”
Put differently: the convenience of automatic inheritance is something GitHub is already considering taking back. Admins who configure their important models explicitly now won’t have to do this twice. This is at the feedback stage — the discussion lives in GitHub Community, and no timeline has been published.
6. September 1 isn’t only about this
The same date carries a second change for Copilot’s business customers. This site previously verified that GitHub’s transitional AI Credits bonus for existing Business and Enterprise customers expires 2026-09-01 (Enterprise drops from 7,000 back to 3,900 points per seat; Business from 3,000 back to 1,900) — see the Copilot 9/1 credit cliff check.
The practical upshot of both landing together: models auto-enabling (people may reach for pricier ones) plus a bonus expiring (fewer credits to spend) means September’s usage bill can look nothing like August’s. To estimate what different models consume, see this site’s Copilot model multiplier breakdown.
7. What this site could not verify — and therefore did not write
Per this site’s standard, none of the following is stated as fact, because no official source backs it:
- Which day your specific enterprise gets the rollout. GitHub says only “gradually, through September 1” and that timing differs per enterprise. No schedule, no lookup method.
- Whether admins get notified when the policy takes effect. Neither the changelog nor either docs page mentions a notification.
- Whether auto-enabled models change your bill. Plausibly related to model multipliers, but this announcement says nothing about billing, and this site won’t infer it.
- Whether individual plans (Free/Pro/Pro+/Max) will get something similar. The official scope is Business and Enterprise; nothing is said about individual plans.
- When “Delegate to default policy” might be removed. GitHub says “evaluating.” No date.
Official sources
- Global model policy generally available — GitHub Changelog, 2026-08-26
- About default availability of Copilot models — GitHub Docs
- Managing availability of models in your enterprise — GitHub Docs
Related checks on this site
- GitHub Copilot tool page (plans and free tier, verified)
- Copilot’s 9/1 AI Credits cliff, verified
- Copilot model point multipliers
- Copilot free vs paid: what actually differs
Every fact above comes from the official pages listed, verified 2026-08-27. The rollout is staged — your enterprise’s actual state is whatever the admin console shows.
What Amo and Pimi think
Students: Pro is free after certification, no reason not to use it. General developers: Free version with 2,000 completions to get started, upgrade to Pro for US$10/month after writing every day. However, it's a fact that free and student plans can only use Auto-select models, so if you mind, you can pay for it.
Let's take a look at these
- GitHub Copilot Comprehensive Introduction: Pricing, Features, and Actual Limitations
- GitHub Copilot Is the free quota enough?
- GitHub Copilot Alternatives
- Comprehensive Free Quota List for All Tools

