AI policy template for game studios: one page your team will follow
For leads and producers whose team already uses AI assistants, with or without rules. This guide walks through the six questions an AI policy has to answer, gives a one-page template you can copy, and shows how to set those rules in Gameheim for the AI that goes through it.
Why most teams use AI without rules
Your developers are probably using AI already. The question is whether the studio has decided how.
The 2026 Stack Overflow Developer Survey asked developers how AI tools are used at their organization, one answer each (13,167 answers). The most common was "AI use is mostly optional and left up to individuals", at 29.7%. "My organization provides approved AI tools and guidance for how to use them" was chosen by 24.0%.
The same survey asked how much developers trust the output of AI tools in their workflow (14,304 answers). 48.0% answered "I trust it when I can easily verify the output". 6.6% answered "I trust it for many tasks, including important work decisions".
Trust, in other words, is conditional: almost half of these developers rely on AI only when they can check its work, and very few trust it with important decisions. Yet for almost a third of them, how AI is used is left to each person. Where that is the case, nobody decides which tools are used, what data goes into them, or who checks the result.
Without a shared rule, each person decides alone:
- Which tool. One developer pastes code into a chat tool, another uses an assistant in the IDE, a third uses none. Nobody knows which company receives what.
- Which data. Unreleased story, platform SDK code covered by an agreement, player data from a playtest: each person draws the line in a different place.
- Who checks. An AI-written changelist description or bug analysis goes in as it is, because checking it was nobody's job.
- No trace. When something goes wrong, there is no way to tell whether a change came from a person or from an assistant acting for them.
A policy does not need to be long. It needs to answer six questions, and the answers need to be set in your tools, where people work.
The six questions an AI policy answers
| Question | What to decide |
|---|---|
| 1. Which tools? | The assistants the studio approves, and for which work. |
| 2. Who may use them? | Everyone, some roles, or people who ask. |
| 3. What data may they see? | Projects, files and information that may reach an AI provider, and what never may. |
| 4. What may AI do on its own? | Drafting and explaining, or also changing, submitting, approving. |
| 5. Who checks the output? | Who reads what AI wrote before it counts. |
| 6. What is recorded? | How you can tell later what was done by an assistant. |
Gameheim can't see what people paste into tools outside it. What it does is give the approved route a home: the AI the studio allows, with permissions, project limits and a record. When the approved way is the easy way, fewer people go around it.
How to write an AI policy and apply it in Gameheim
- The LEAD role, or a role with
roles:manage, to grant permissions in roles. - A plan that includes AI, for the AI features built into Gameheim.
Decide which tools are approved
Write the list down, with what each tool is for: a coding assistant in the IDE, a chat assistant for documents, the AI built into your production tools. A short list people can follow beats a long list nobody reads.
In Gameheim there are two kinds of AI:
- Built-in AI: the writing assistant in the wiki, changelist descriptions drafted from your template, and the explanation of Unreal crash callstacks. It runs entirely on Microsoft Azure, in European data centers. What your team sends to it, and what it writes back, is never used to train AI models, ours or anyone else's: it is used only to provide the service.
- Your own assistants: Claude, ChatGPT, Cursor or VS Code, connected to Gameheim through its MCP server and running on each person's own subscription. See the AI assistants page.
Decide who may use them
Open Roles & Permissions. Two permissions matter here:
Permission Allows In the preset roles ai:use(group AI)Asking the built-in AI to write: the wiki assistant, changelist descriptions, Explain this crash None External apps (MCP & API) Connecting an assistant or a script that acts for the person None. A VIEWER can never have it. Add them to the roles that need them, or create a role for the people who do. See roles and permissions.
One AI feature runs without anyone asking: the explanation of a new Unreal crash. It is on by default; switch it off for a project in Bugs > Settings > Unreal crashes, with Explain new crashes with AI automatically. Explain this crash keeps working for people with
ai:use.Decide what data may reach an AI provider
Write down what must never be pasted into an AI tool, for example content covered by a platform or publisher agreement. Check those agreements: some say where code or builds may go.
Then set it where you can:
- Connected assistants: when someone connects an assistant, they choose the projects it can reach and can untick the permissions it does not need. What a tool returns becomes part of the conversation and reaches the assistant's provider under that person's agreement with them.
- Crash explanations: the model receives the structured fields of the report (crash type, error message, the first 40 lines of the callstack, versions, platform). The minidump, the game log and the machine id are never sent. See crash grouping and triage.
Decide what AI may do on its own
The simplest rule: AI drafts, a person decides. In Gameheim this is how AI works:
- A changelist description drafted by AI is reviewed before submit.
- An assistant connected through MCP can do only what its user can do, and nothing can be deleted through Gameheim's tools.
- Changes to the shared instructions of your team's assistants are proposed by the assistant and approved by a person, always in Gameheim. See shared AI memory for game teams.
Decide who checks the output
Name the role that reads AI output before it counts: the reviewer of a changelist, the QA lead for a crash analysis, the owner of a wiki page. The crash explanation in Gameheim is marked as generated by AI and possibly wrong: treat it as a first read, to confirm in the debugger.
Decide what is recorded, and review the connections
- In Audit Log, a change made through an assistant shows the app next to the person, for example via Claude.
- In Integrations, AI assistants, a LEAD sees the apps connected in the organization and can disconnect them. A connection closes by itself after 90 days without use, and after one year in any case.
A one-page AI policy template
Copy it, fill in the brackets, keep it on your wiki.
AI policy templateAI policy: [Studio name], version [date]
1. Approved tools
[Tool]: for [work]. [Tool]: for [work].
Other AI tools are not used on studio work.
2. Who may use them
[Roles] may use the built-in AI. [Roles] may connect their own assistant.
3. Data
Never shared with an AI tool: [content under NDA, platform SDKs, player data, ...].
Connected assistants reach only these projects: [projects].
4. What AI does on its own
AI drafts, explains and answers. A person submits, approves and decides.
5. Who checks
[Role] reviews [AI output] before it is used.
6. Records and review
Changes made through assistants are in the audit log.
This policy is reviewed every [6 months] by [owner].Frequently asked questions
Do we need an AI policy if the team is small?
Yes, and a short one. Six answers on one page are enough: which tools, who, which data, what AI may do alone, who checks, what is recorded.
Can my team use AI in Gameheim without me knowing?
Asking the built-in AI about your project's content and connecting an assistant both need a permission that only a role you control can grant, and a LEAD sees every connected app and can disconnect it. The only AI that runs by itself, the explanation of new Unreal crashes, can be switched off for each project in the bug settings. AI used outside Gameheim is not visible to it: that is what the written policy is for.
Does a connected assistant see every project?
No. Its user chooses the projects when connecting, and it can do only what its user can do.
Where does the built-in AI run?
On Microsoft Azure, in European data centers.
Can an AI assistant delete data in Gameheim?
No. Nothing can be deleted through Gameheim's tools.