Responsible AI Policy

1. Scope

This policy covers the AI features of the DXAPlay platform: the AI assistants in the back office (an analytics assistant, a segment builder and an SEO keyword helper) and the role-specific agents of the AI Workforce.

2. Principles

• People remain accountable. Every agent has a named human owner, and every consequential action has a named human approver.
• AI assists, never replaces, mandated roles. Agents assist the MLRO, the compliance officer, legal counsel and responsible-gaming professionals; they never replace them, and hiring agents does not reduce an operator’s duty to staff those functions.
• Player protection first. Responsible-gaming and anti-money-laundering considerations override commercial objectives.
• Least privilege and least data. Agents get only the tools and data their task needs, through the same permission checks as people.
• Tighten, never loosen. Agents may restrict, pause or escalate within their permissions; lifting a control is for people.
• Fail closed. When a safety check cannot answer, the agent does not act.
• Transparency. AI-generated content is labelled, and players are told when they interact with AI.

3. Decisions reserved for people

AI never makes final decisions on player account closure, AML or suspicious-activity determinations, regulatory reporting, large balance adjustments, withdrawal rejections, responsible-gaming interventions, legal determinations, changes to financial controls, production deployments, high-value promotional liability, contract execution, employee termination or material pricing changes. Agents may prepare evidence and drafts for the person who decides. The full list is on the AI Workforce page.

4. Data use

• Each operator’s data is used only for that operator. Agents cannot access another operator’s data.
• DXAPlay does not use customer or player data to train shared models unless an explicit, legally reviewed agreement with the customer permits it.
• AI features send the minimum data needed to the model provider listed on the sub-processor list.

5. Oversight, records and limits

Agents are designed to start in shadow mode, to move to more autonomy only with evidence and the human owner’s sign-off, to record every run (inputs, data used, output, approval and cost), and to stop at a kill switch or cost cap. AI outputs can be wrong; DXAPlay does not promise their accuracy and commits instead to evaluation, human approval and monitoring.

6. Questions under legal review

Counsel is reviewing, among other things, how the EU AI Act and automated-decision rules apply to AI agents and to operators that use them.

7. Contact

For questions or concerns about DXAPlay’s use of AI, use the contact form and choose the topic “Legal”.