What it does

The org and enterprise controls for running Copilot safely: policies, content exclusions, audit logs, duplication detection and IP indemnity, data-handling terms, SSO and seat management. All of it lives on Business and Enterprise plans.

Key facts

• Admins set Copilot policies at org/enterprise level — which features are on (chat, coding agent, CLI), and enterprise policies can be inherited by orgs.
• Content exclusion blocks named files, directories and repos; excluded paths aren't sent to Copilot. Watch this one: agent mode and cloud agents have not respected exclusions (as of May 2026).
• Audit logs record Copilot admin events (access changes, policy edits) and retain events for 180 days; GitHub recommends streaming to a SIEM for longer history.
• Duplication detection filters verbatim and near-verbatim matches against known public code and can flag the source licence — it does not catch paraphrased or partial rewrites.
• IP indemnity is offered to Business and Enterprise customers, but only applies when duplication detection is enabled.
• On Business/Enterprise, prompts and suggestions are not retained and are not used to train models; user engagement data is kept for around two years.
• Seat management runs through GitHub org membership, teams and direct assignment; SSO/SAML gates access via the GitHub Enterprise Cloud identity setup.
• Coding agent is off until an admin enables it on the Policies page.

When to use / skip

Any corporate rollout goes on Business or Enterprise — full stop. The individual plans lack the policy plane, the data terms and the indemnity, so they are for individuals only. Turn duplication detection on and keep it on; the IP indemnity is worthless without it, and the cost is negligible. Use content exclusion for secrets and sensitive paths, but don't treat it as a hard security boundary given the agent-mode gap.

Configuration decisions

• Decide the policy baseline: which features enabled, set at enterprise level and inherited, or per-org.
• Turn duplication detection on org-wide — it's the precondition for IP indemnity.
• Configure content exclusion for sensitive repos and paths, and audit that it's actually applied where it matters.
• Set up audit log streaming to your SIEM; 180 days in-product is not enough for most compliance regimes.
• Wire SSO/SAML and decide seat assignment model — automatic for all members, or explicit by team.

Gotchas

• The content-exclusion gap for agent mode and cloud agents is the big one. Excluded paths can still reach those surfaces — verify current behaviour before you certify it to security.
• IP indemnity silently doesn't apply if duplication detection is off. Check the setting, don't assume.
• Audit retention is only 180 days in-product; without SIEM streaming, older evidence is gone.
• Duplication detection only catches near-verbatim copies — it won't flag reworded or structurally similar code.
• Engagement data retention (~2 years) is separate from the "no training / no prompt retention" guarantee; know the difference when answering DPO questions.

Consultant notes

• Lead security and legal conversations with three things: no-training data terms, duplication detection plus indemnity, and audit/SIEM streaming. That covers most procurement checklists.
• Be straight about the agent-mode content-exclusion gap — it's the caveat that catches careful clients, and hiding it costs trust.
• Map Copilot governance onto the existing GitHub Enterprise Cloud identity and audit setup rather than treating it as a new silo.
• Phase rollout with policies: completions and chat first, coding agent and CLI once data-flow questions are answered and signed off.

---

Verify duplication detection is actually enabled in the client's org — the indemnity depends on it and it's easy to miss.