Massachusetts Already Regulates Sportsbook AI. The Rule Turns on What DraftKings Knew.

On September 24, the Massachusetts Gaming Commission said it will review how its licensed sportsbooks use artificial intelligence and machine learning, starting with DraftKings. The trigger was a New York Times investigation published the previous Saturday.

The Times reported that DraftKings directed employees to build a machine-learning model to identify customers most likely to respond to promotions and then lose money, including an "elasticity" metric estimating how much more a customer would wager after an offer. The paper said its reporting drew on more than 40 former employees, internal research memos and Slack messages, and customers' betting records. Six former employees who worked on the systems told the Times they feared the technology could harm people already struggling with gambling. The Times also reported that work on similar technology to identify gamblers who might need help had been sidelined.

Chair Jordan Maynard said he had asked Executive Director Dean Serpa and staff to engage with DraftKings to understand the specifics, and that the review would extend to every licensed operator. The Commission has made no finding that DraftKings did anything wrong. DraftKings said it works to comply with all Massachusetts regulations and denies doing anything to make its offerings addictive.

Massachusetts is unusual here, and the reason is one regulation. Commissioner Paul Brodeur said at the meeting that Massachusetts is one of the only states, if not the only one, to reference AI in its gaming rules.

The rule is 205 CMR 257.02, part of the Commission's sports wagering data privacy regulations. It lists factors an operator may not use patron data to promote wagers or offers on. One of them is any computerized algorithm, automated decision-making, machine learning, artificial intelligence or similar system that is known or reasonably expected by the operator, or its vendor, to make the gaming platform more addictive. The same list bars targeting promotions based on a patron's use of responsible-gaming tools: the limits they set, how often they change them, and whether and how often they take cooling-off periods.

A separate provision governs consent. An operator that wants to use patron data beyond the legitimate business purposes the rule lists has to get consent that is clear, conspicuous and obtained apart from any other agreement, and acceptance of general terms of use does not count.

The Commission's authority for all of this comes from the state's sports wagering statute, Chapter 23N, which is also the source of its power over licensees.

Read closely, the AI provision does not ban what most coverage describes. It does not prohibit using machine learning to target promotions, and it does not prohibit targeting customers who are likely to lose. Every sportsbook's customers are, on average, likely to lose. The prohibition is narrower and sharper: a system the operator knows, or reasonably expects, makes the platform more addictive.

That puts the entire review on a question of knowledge, and it is why the Times's reporting matters legally rather than just reputationally. The paper did not only report what the model did. It reported what people inside the company said about it: that former employees who built the systems worried it could harm people already struggling, and that a parallel effort to identify at-risk gamblers was set aside. If those accounts hold up, they are the kind of evidence that speaks to what an operator reasonably expected. A regulator asking what DraftKings knew now has a list of people to ask.

There are real limits on that reading, and they cut in DraftKings' favor. "More addictive" is a demanding standard, and a model that predicts who will respond to an offer is not self-evidently a model that makes a platform addictive. The rule's wording also ties the prohibition to promotions targeted on these factors, which leaves room to argue about what the model was actually used for in Massachusetts specifically. The Times's reporting is national. The Commission's jurisdiction is not.

The responsible-gaming provisions may prove easier to apply. They bar targeting based on how a patron uses limits and cooling-off tools, which is concrete data rather than an expectation. If any Massachusetts promotion model drew on those signals, the knowledge question never arises.

None of this is a finding. The Commission has announced a staff review, not an adjudicatory proceeding, and it said further action or policymaking would come only if and when appropriate.

The Commission, for now, and it has more than it is using. A staff review costs it nothing and lets it gather facts before choosing between enforcement against one operator and new rules for all of them. DraftKings has the facts: it knows what its models used as inputs in Massachusetts, what employees said about them, and when. That is leverage in the sense that the review runs on its documents, and exposure in the same sense. The other operators are in the review too, which gives them every incentive to show their own systems stay clear of the rule's factors before anyone asks. And the former employees quoted by the Times hold something no one else does: firsthand knowledge of what was expected, which is the element the rule turns on.

Every sportsbook targets customers likely to lose. Massachusetts only bans it when the operator knows the model makes betting more addictive. The case is about what DraftKings knew.

Watch for the Commission's next public meeting agenda for any report back from staff, and for whether the review becomes a formal adjudicatory matter against DraftKings. Watch too for whether the Commission's AI task force turns this into rulemaking that sharpens "known or reasonably expected," and whether any other state copies the Massachusetts provision now that it has a live test.