Massachusetts Probes and Baltimore Sues Over DraftKings AI Promotions

Wednesday 30 September 2026 topic: how losing-gambler targeting claims turned into live probes and lawsuits

This is a decorative lead image for an editorial piece with no numeric data to plot — the brief explicitly wants tone and tension conveyed conceptually, not a literal news…

For years the polite fiction was that the bookie just took your bet. This week that fiction copped a battering in two places at once. Massachusetts is probing how DraftKings used AI to target losing gamblers, with its gaming commission set to examine how DraftKings and other licensed sportsbooks deployed artificial intelligence in promotions.1 At the same time, the City of Baltimore has lodged an action in its circuit court against DraftKings Inc and Flutter Entertainment trading as FanDuel.2 I have watched enough of these gambling-tech scandals to know the pattern: exposé, denial, shrug. This feels different, because the paperwork has started.

The detail behind the outrage is what made me sit up. The machine-learning model DraftKings built in 2023 scanned playing frequency, daily balance patterns and the ratio of losses to total wagers, then gave each casino player a score predicting how much they would lose after a free bet or bonus.3 That is not spray-and-pray marketing, that is elasticity modelling: find the punter most likely to lose, then nudge him to play. One account summed up the commercial logic as “The best investment would be a problem gambler.”4 DraftKings, for its part, pushes back hard. An executive is reported via the New York Times coverage as saying leaders made a decision “not to use predictive technology for problem gambling”.5 I read that as a careful denial, and it leaves the key question hanging: if the loss-prediction score was only about promotion efficiency, why build it around loss behaviour at all.

Here is where I have to pump the brakes, because the legal path is messier than the headlines suggest. A federal court has just ruled that sports bets on prediction markets are swaps exempt from state gambling laws, with the court holding that Washington preempts states applying their own gambling rules there.6 That is a different product category, prediction markets rather than licensed sportsbooks, but it shows how quickly state-level action can be outflanked when a judge decides gambling is finance. And the blokes in the forums are already cynical about regulators catching up. One commenter put the industry view bluntly: the lucrative customers and the addicted gamblers are “These are the same people.”7 Another called the apps a “bleeding edge addiction platform” where ordinary punters do not stand a chance.8 I take that as sentiment, not proof, but it captures a real limitation in the enforcement story: the same data that identifies harm also identifies value.

My take is that the response is real but narrow, and that narrowness is the point. Massachusetts has not banned anything yet, it has opened a review. Baltimore has not won anything yet, it has filed. Yet compared with the usual cycle of think pieces, that is concrete shape: investigators asking for models and messages, lawyers testing duty and deception theories in court. I do not think either action will kill behavioural targeting on its own, especially while federal preemption arguments hover in the background. What they do is force discovery, and discovery is what this industry hates. Once the scores, the thresholds and the promo triggers are on the record, the argument stops being about AI ethics in the abstract and becomes about whether a licensed operator knowingly fed bonuses to people its own maths said would lose. That, to me, is where this finally gets interesting.

Sources

How this was made
  • 01-research ~z-ai/glm-flash-latest $0.009
  • 04-nominate ~deepseek/deepseek-v4-flash-latest $0.022
  • 05-select ~google/gemini-flash-latest $0.008
  • 06-write meta/muse-spark-1.3-contributor $0.002
  • 08-visualise anthropic/claude-sonnet-5 $0.040

total $0.081

What each stage does, drawn out →