AI operators in Massachusetts face questions they cannot answer

Massachusetts AI questions expose what operators cannot prove about their own systems

The Massachusetts Gaming Commission sent licensed sportsbook operators a list of more than 70 questions on October 14. The answers are due November 3. Participation is voluntary. That last detail is worth holding.

The review began with a New York Times report alleging that DraftKings used machine learning to identify losing bettors and route promotions toward them. DraftKings denied wrongdoing. The MGC, to its credit, did not treat this as a DraftKings problem. It treated it as an industry problem, which is the correct read, and it sent the questionnaire to every licensed operator in the state.

The questions are careful in a way that reveals what regulators have already concluded. The MGC's definition of AI covers any automated tool that ranks, classifies, or makes predictions about patrons — including systems operators describe as "personalization," "optimization," or "risk assessment." That definitional sweep closes the most obvious escape route before operators have a chance to use it.

The section that matters most is not the one asking whether losing bettors are treated differently. It is the one that follows. Operators who answer no are asked to explain the controls that prevent it. That is not a follow-up question. That is the question. An operator that cannot describe the control does not have one.

The final section asks whether a human reviews a model's recommendation before it goes out, whether that person can override it, and whether they receive an explanation of why the model flagged a particular patron. Then the MGC asks something more precise: if a regulator asked why Patron A received Promotion X on a specific day, could the operator reconstruct the answer? Could the operator calculate what percentage of that patron's promotions were AI-influenced?

I have sat across the table from compliance officers who believed, in good faith, that their systems were auditable. The belief and the audit are different things. The MGC is not asking whether operators believe their systems are explainable. It is asking whether they can explain them, to a specific question, about a specific person, on a specific day.

Michigan has said it is reviewing the same Times report. Its gaming laws and regulations do not reference AI. That gap is not incidental — it means Michigan lacks the enforcement hook Massachusetts has already written into its rules. MGC Commissioner Paul Brodeur noted that Massachusetts may be the only state to have AI explicitly in regulation. That head start is now an enforcement advantage.

The MGC has not announced further action. Chair Jordan Maynard said that any policy steps would follow review of the responses. Massachusetts law already prohibits AI that is "known or reasonably expected" to make a gaming platform more addictive. The standard is not intent. It is what the operator knew or reasonably should have known about its own system's effect.