Player agency
Intelligence should expand useful choices, not quietly take control away from people.
Design adaptations with clear boundaries, appropriate control, and a way for people to understand what the system is doing.
AI principles
Build intelligence around people, purpose, and play.
Games are human systems as much as technical ones. We believe intelligent systems should be useful in context, understandable to their audience, and designed to preserve player agency and human judgment.
Our working commitments
Principles that shape system decisions.
These principles are a public design posture. The controls, evaluation, and disclosures needed for a particular system still depend on its audience, consequence, and deployment context.
Purpose limitation
Gameplay data should be used for an explicit, understandable, and legitimate purpose.
Define the decision or experience a system supports, then prevent unrelated uses from becoming the default.
Data minimization
More data is not automatically better understanding.
Collect and retain only the signals needed for the stated purpose, with boundaries suited to the deployment.
Audience-appropriate transparency
People need explanations they can use—not technical detail for its own sake.
Make system purpose, important limitations, and the basis of consequential outputs understandable to each audience.
Human oversight
People remain accountable for decisions that require judgment, context, or care.
Create review, escalation, and override paths proportionate to the consequence of the system output.
Security and abuse resistance
A useful intelligence system must anticipate misuse as well as ordinary failure.
Apply access boundaries, input validation, monitoring, and misuse analysis appropriate to the public surface.
Evaluation and monitoring
A system should be evaluated in context before launch and observed after conditions change.
Test usefulness and failure modes, document assumptions, and establish a response when performance or impact drifts.
Publication boundary
Rights, provenance, and training data
Specific public claims about data rights, provenance, and model training remain unpublished until the relevant practices and language are finalized.
Do not infer a training-data policy or rights position from the absence of a claim on this website.
From principle to practice
Questions we want every system to answer.
Principles matter when they change the questions a team asks before launch, during evaluation, and when conditions evolve.
- 01
What purpose does the system serve, and who benefits?
- 02
What is the minimum information needed to serve that purpose?
- 03
What can the system get wrong, and who notices?
- 04
Which choices must remain with a player or human decision-maker?
- 05
How will limitations and important outputs be explained?
- 06
What monitoring, escalation, and retirement path does the system need?
Questions or feedback
Responsible practice improves through scrutiny.
If you have a question about these principles or how they should apply to our work, we welcome the conversation.