Massachusetts AI probe turns on a question operators cannot answer

Dean Sherpa sent a five-page letter on October 9 to every licensed mobile sports betting operator in Massachusetts. The deadline is November 3. The letter runs to more than seventy individual questions. The one that matters most is also the simplest: if a regulator asked why a specific patron received a specific promotion on a specific day, could the operator identify the data, the model, and the factors that produced that recommendation?

The Massachusetts Gaming Commission already knows some operators cannot. That is why the letter exists.

The review was triggered by a New York Times report alleging that DraftKings used machine learning to identify bettors most likely to lose and then targeted them with promotions designed to keep them playing. DraftKings denied wrongdoing. MGC Chair Jordan Maynard announced in September that the commission would examine all seven licensed operators, not only DraftKings, on the grounds that the practice, if it exists, is almost certainly not unique to one company.

What Sherpa's letter reveals is that the commission has moved past the question of intent. Whether an operator designed a system to prey on losing bettors is difficult to establish. Whether an operator can explain what its system is doing is a much simpler test, and far easier to fail.

The letter defines AI broadly enough to catch almost anything: any automated tool that ranks, classifies, or makes predictions about patrons qualifies, regardless of whether the operator calls it personalization, optimization, or risk assessment. Systems developed internally and systems bought from third-party vendors are both in scope. The final section of questions asks specifically about human oversight — at what point a person reviews a model's output, whether that person can override it, and whether the employee receives any explanation of why the model made a particular recommendation.

That last detail is where the regulatory exposure concentrates. An operator that cannot tell a regulator why Patron A received Promotion X has a structural problem that predates any finding of harm. Massachusetts already bars the use of AI reasonably expected to increase addictive behavior, which means the question of explainability is not academic — it is the threshold question for whether existing rules have been broken.

Michigan's Gaming Control Board told reporters it is reviewing the issues raised in the Times report, though it stopped short of committing to a formal inquiry. Michigan's statutes do not reference AI. That gap is significant: a Massachusetts finding that operators cannot explain their own systems would land in a state with no equivalent regulatory hook, and the enforcement options would be materially different.

The consensus reading of this process is that it ends with disclosure requirements and perhaps some enhanced oversight language. The consensus is probably right about the outcome but wrong about the mechanism. The disclosure requirements, when they come, will be written around the explainability standard Sherpa has already established. Operators that built systems they cannot interpret will not be able to comply by writing a policy document. They will have to rebuild the systems or decommission them. That is a different order of cost than a fine.