Massachusetts AI review exposes a gap operators cannot close

Paul Brodeur said it plainly last month: Massachusetts is one of the only states in the country to reference AI in its gaming regulations. That sentence deserves more attention than it has received, because what it describes is not a regulatory advantage. It is a measure of how far behind everyone else is.

The Massachusetts Gaming Commission sent more than seventy individual questions to every licensed sports betting operator in the state, with responses due by November 3. The review is voluntary. Responses will be kept confidential. Those two facts are doing a lot of work, and not in the regulator's favor.

The questions themselves are the story. The MGC wants to know whether operators can identify, for a specific patron on a specific day, which data inputs caused a model to recommend a particular promotion. It wants to know whether a human reviewed that recommendation before it went out, and whether that person could override it. It wants to know what percentage of any given patron's promotional activity was shaped by machine learning rather than a human decision.

These are not aggressive questions. They are basic questions. The kind a compliance officer at a bank would answer before lunch. The fact that the MGC is asking them to an industry that has been operating in Massachusetts for years, and asking voluntarily with confidential responses, tells you something about what the regulators expect to find when the answers come back.

The review was triggered by reporting that DraftKings used AI to identify losing bettors and target them with promotions. DraftKings denied wrongdoing. What the MGC understood — and what matters here — is that the specific company is not the point. If one operator built a system capable of that targeting, the architecture is available to all of them. The question the commission is now asking every licensed operator is whether they built it too, or whether they could even tell.

I have watched regulators try to audit systems that operators themselves could not fully explain. The standard response is not deception — it is genuine opacity. The model was trained, it produced outputs, the outputs were acted on, and the chain of human decision-making that would satisfy a regulator was never built because no one required it. Massachusetts is now requiring it, or at least asking whether it exists.

MGC Chair Jordan Maynard was careful last month: any further action would be determined "when and if appropriate." That hedge is doing what hedges do. But the questions themselves have already moved the line. An operator that cannot answer why Patron A received Promotion X on a given day has a different relationship with its own compliance program than it had before this review began. Whether the MGC has the statutory reach to act on what it learns is a separate problem. The disclosure requirement, even voluntary, changes what operators have to be prepared to say out loud.

Michigan is watching. It has not committed to a DraftKings-specific review, and its regulations do not address AI. That gap is now visible in a way it was not six weeks ago.

The operators who cannot answer the MGC's questions are not necessarily doing something illegal. They may simply have built systems that outran the accountability structures around them. In most industries, that condition resolves quietly. In a regulated gambling market, it resolves in front of a commission that has already said it will decide what to do with the answers.