Here is a story that should end a few careers. According to a New York Times investigation published in mid-September, the sports-betting giant DraftKings built a machine-learning model that scores every single customer on how much money they are likely to lose, and then aims its biggest “free money” bonus bets at the people the model predicts will lose the most. Not the most engaged customers. Not the biggest fans. The biggest losers. The model reportedly had a name for the score internally: “elasticity.” A higher number meant you were worth chasing with more perks, because you were more likely to keep bleeding.
The part that turns a story into a scandal
Plenty of companies use AI to find their best customers. What makes this genuinely damning is the road not taken. The Times reports that a DraftKings data scientist, Nestor Hernandez, started building a different model in 2024, one designed to flag gamblers sliding toward a crisis so the company could intervene. That project was shelved. According to the reporting, DraftKings’ own Chief Responsible Gaming Officer said leaders reached a “collective decision” against predictive tools of that kind, on the grounds that the approach was not “evidence-based.” So a predictive model to find who to protect: not evidence-based, canned. A predictive model to find who to exploit: built, and credited with a 13% margin boost in 2025. The maths of that decision tells you everything about which way the company pointed its cleverness.
DraftKings denies the framing, telling the Times it “rejects any implication” that its marketing unfairly targets customers, and says promotions go to users who show “sustained, engaged” use rather than people ranked by how much they lose. You can weigh that against the reporting yourself.
Why this matters beyond one betting app
This is the clearest, ugliest example yet of the thing critics keep warning about: AI does not have to be superintelligent to do harm, it just has to be very good at optimising for a metric that happens to hurt people. An “elasticity” score is not science fiction. It is a spreadsheet with a conscience surgically removed, and the same technique works for any business with addictive potential and a data warehouse, from gambling to gaming to trading apps. The lesson for regulators is that “the algorithm did it” is going to be the defence of the decade, and the useful question is never what the model does but what metric its owners told it to chase. Point a capable AI at “maximise losses from the vulnerable” and it will oblige, cheerfully, at scale, until someone with subpoena power makes it stop. (Sources: New York Times investigation, September 2026.)
Related: when an algorithm makes the call, who’s responsible?