Massachusetts gaming regulator expands AI investigation to all sportsbooks

Massachusetts Gaming Commission widens AI inquiry to all licensed sportsbooks

Jordan Maynard said it in plain language: the Massachusetts Gaming Commission would engage with DraftKings to "understand the specifics" of its AI and machine-learning practices. That phrasing — careful, conditional, not yet accusatory — is the language of a regulator who has read something alarming and is not yet sure what they are holding.

What they are holding is a New York Times investigation reporting that DraftKings used machine learning to identify customers most likely to keep gambling after receiving promotional offers. The system, according to the Times's account based on interviews with more than forty former employees, also applied an "elasticity" metric to estimate how much more a given customer might wager following a promotion. Six of those former employees told the Times they believed the technology could harm people struggling with addiction. Efforts to use similar modeling to flag customers who might need intervention, the Times reported, had been sidelined.

DraftKings denied the characterization. The company said it does not use AI to target customers based on losses and does not market to customers on the basis of indicators of problem gambling.

The Commission's inquiry will not stop at DraftKings. Maynard confirmed it would extend to all licensed sportsbooks operating in the state. That is the sentence that matters most for anyone trying to read where this goes.

Massachusetts law already prohibits using customer data to provide promotions through automated systems — including AI — that are "known or reasonably expected" to make a gambling platform more addictive. Operators must report at least twice annually on how they analyze customer behavior to promote responsible gambling. The Commission has not previously taken enforcement action on an AI-related matter. That prior restraint is not a safe harbor. It means the enforcement record is blank, and the first case written onto it will define what the statute reaches.

The distinction the Commission will have to draw — and that any enforcement action would turn on — is between optimization that responds to customer preference and optimization that exploits behavioral vulnerability. Those are not the same thing, but they can look identical from the outside, and the data needed to tell them apart sits inside the operator's systems, not the regulator's office.

I have seen regulators in adjacent contexts treat the absence of prior enforcement as evidence that the conduct was permissible. It is not. It is evidence that no one had looked yet. The Commission is looking now, and its mandate gives it the tools: fines, license suspension, revocation.

The standard Massachusetts regulators will apply is whether the AI system was "known or reasonably expected" to increase addictive behavior. That is an objective foreseeability test, not an intent test. What the engineers believed about their system's effects is relevant. What the system actually did is more relevant. What six of them told a major newspaper about their concerns is, at minimum, notice.