Can AI stop gambling addiction before damage occurs? 

A new algorithm can assess which online gamblers face an elevated risk based on actual betting behavior. The open-source model, developed by researchers at the University of Amsterdam (UvA), looks beyond just the amount money a player loses. Bet patterns, playing times, and responses to winning and losing streaks are also taken into account. 

This brings an uncomfortable question increasingly to the fore: what happens when an online casino can detect that gambling is spiraling out of control even before the player realizes it? A technical signal only carries value if it leads to a warning, personal contact, an intervention limit, or appropriate assistance. 

Online gambling leaves a highly detailed digital footprint. An operator can see when a user logs in, how long a session lasts, how much money is deposited, which games are selected, and how quickly a player bets again after a loss. In isolation, these data points may reveal little. Together, they can highlight behavioral shifts that the player still denies or fails to recognize. 

Two years of bets across thirteen casinos analyzed 

The model was developed by UvA PhD candidate Charles de Leau alongside professors Reinout Wiers and Johan Bollen. According to the University of Amsterdam, the algorithm was trained on all player bets across thirteen Dutch online casinos over a two-year period, spanning from July 30, 2023, through July 30, 2025. 

As a result, the study does not rely on survey responses or players’ retrospective memories of their behavior—it utilizes actual recorded wagers. This provided researchers with the opportunity to identify behavioral patterns that precede severe problem gambling. 

The machine learning model evaluates bet sizes, wager frequency, session timing (such as playing late at night across consecutive days), and responses to win or loss streaks to generate a risk score. This score is not a clinical diagnosis and does not prove an active gambling addiction; rather, it serves as an indicator that specific behavior warrants closer attention. The Netherlands Gaming Authority (Ksa) emphasizes that the model is merely a supplementary tool and that operators cannot rely solely on an algorithm to demonstrate compliance with their duty of care. 

Behavioral shifts matter more than single deposits 

Problem gambling behavior can rarely be captured by a single fixed threshold. A €500 deposit might represent responsible play for one individual, while for another, it could mean being unable to pay rent. Similarly, a lengthy playing session does not automatically indicate an addiction. 

Risks become clearer when multiple warning signs converge: a player bets more frequently, raises stake sizes, redeposits shortly after a loss, extends session times into the night, takes fewer breaks, and attempts to chase losses at an accelerating pace. 

An algorithm can be far more effective at spotting these nuanced shifts than simple fixed limits. By comparing behavior over an extended period, the system identifies deviations from a player’s personal baseline. Gambling Club previously detailed why monetary velocity, playing time, and wagering frequency serve as crucial red flags for gambling addiction. 

While a player might view their choices as a sequence of isolated decisions—one more deposit, one more round, or one last attempt to break even—a data-driven system analyzes the aggregated pattern, making it harder to dismiss systematic behavioral changes as mere coincidences. 

France also deploys risk-detection algorithms 

The Netherlands is not alone in pursuing this approach. In France, the Autorité nationale des jeux (ANJ) utilizes technology to better identify players displaying potentially excessive gambling habits. Gambling Club previously reported on how an algorithm deployed by the French regulator revealed the true scale of excessive gambling. 

The Dutch model was partly inspired by work conducted by the Spanish regulator, the DGOJ, demonstrating that data-driven player protection is gaining momentum internationally. Regulators across Europe are seeking ways to transform vast streams of transactional data into actionable, auditable risk indicators. 

The open-source nature of the UvA model is a significant development. Many existing detection tools were created by commercial vendors or gambling operators themselves, keeping their internal mechanics opaque. Because this new model is open source, researchers and regulators can inspect, audit, and refine its code and methodology independently. 

This provides the Ksa with an objective benchmark to compare against operators’ proprietary risk scoring systems, reducing reliance on black-box tools where external oversight cannot easily verify which indicators are weighted or which vulnerable players might be overlooked. 

AI is not an automated solution 

The term “algorithm” can create a false impression that player protection can be fully automated, but the reality is far more complex. Gambling Club has previously highlighted how the gambling industry continues to struggle with the mature and responsible deployment of artificial intelligence, particularly regarding governance, transparency, and specialized expertise. 

Furthermore, models are susceptible to errors. Players may be flagged as high-risk due to benign anomalies, or conversely, individuals experiencing severe harm might go undetected if their habits do not match the specific patterns the system was trained to recognize. 

For instance, a player who deposits large sums from day one exhibits no sudden escalation relative to their past behavior, meaning change-detection models might miss them entirely. The same blind spot applies to individuals who spread their gambling across multiple operators, as each company sees only a fragment of their overall activity. 

Consequently, risk scores must be evaluated alongside supplementary data and human oversight. While the algorithm can flag a file for review, trained personnel must ultimately investigate the situation and engage directly with the player. 

The same data can be used to prolong play 

A fundamental tension exists within the underlying technology: the very behavioral data that enables a casino to detect harm also carries significant commercial value. It reveals game preferences, peak activity windows, and the most persuasive communication channels or promotional triggers. 

While an operator can use these insights to issue budget reminders or suggest cooling-off periods, the same intelligence could theoretically be leveraged to optimize retention and extend play sessions. This renders the boundary between personalized player protection and commercial exploitation exceedingly delicate. 

This dynamic makes strict data transparency essential. Players must be informed about what data is collected, how it is processed, how long it is retained, and whether automated profiling influences account settings or access. Gambling Club previously outlined the specific GDPR obligations imposed on online casinos regarding player data processing. 

Fortunately, privacy and player safety are not mutually exclusive. Pseudonymized datasets can support valuable academic research without exposing individual identities, provided that robust security protocols, clear usage boundaries, and oversight are maintained to prevent sensitive gambling data from being repurposed. 

Detection creates an obligation to act 

Identifying a potential gambling risk is merely the first step; the critical challenge lies in executing appropriate interventions. Automated pop-up warnings are easily dismissed by users. Direct, personal outreach from a trained customer protection representative—referencing specific behavioral changes rather than generic warnings—tends to be far more impactful. 

Depending on the severity of the indicators, an operator may require a player to lower deposit limits, impose temporary deposit freezes, or block account access entirely. When reasonable grounds point to compulsive gambling, assertive intervention becomes mandatory. 

Chasing losses—manifested through escalating stakes, rapid redeposits, and extended sessions—remains a primary indicator of escalating harm. Gambling Club has consistently warned players about the dangers of attempting to recover lost funds. Algorithms are uniquely positioned to catch this behavior early by tracking the speed and scale of bets placed immediately following a loss. 

Advancements in monitoring technology do not lessen an operator’s legal duty of care; they expand it. As detection tools grow more precise, operators will find it increasingly difficult to claim that signs of distress were invisible. At the same time, software should never make automated, life-altering decisions without human review—the quality of human follow-up ultimately dictates the success of player protection efforts. 

Demand for addiction care continues to rise 

The urgency of this issue is reflected in Dutch healthcare data. According to LADIS (National Alcohol and Drug Information System) statistics, over 3,100 individuals received specialized treatment for gambling as their primary diagnosis in 2025. Both the total number of patients and the relative share of gambling-related cases in addiction care have risen steadily year-over-year since 2023, with half of registered clients entering treatment for the first time. 

LADIS notes that this upward trajectory remains consistent even when controlling for reporting consistency among healthcare providers, indicating that the surge is not merely an artifact of expanded data collection. While specific gambling venues are not always fully recorded in clinical intake forms, addiction care providers attribute the recent increase primarily to online gambling. 

An algorithm can serve as an early warning mechanism, but it cannot offer therapy or human empathy. Machine learning models evaluate transactions and behavioral metrics—they cannot perceive personal distress, hidden debts, emotional trauma, or job loss. 

Therefore, early detection technology must bridge digital monitoring and human support systems. Friends and family members are often the first to notice real-world signs of problem gambling, such as uncharacteristic secrecy, unexplained financial strain, or broken commitments. Gambling Club maintains a practical resource guide detailing steps to take if a loved one shows signs of compulsive gambling. 

The UvA’s open-source model provides regulators and operators with a powerful analytical framework, but the true test begins after a risk flag is triggered. Is the alert investigated thoroughly? Does timely, human intervention follow? And is the player effectively protected if harmful behavior continues to escalate? 

Ultimately, technological progress is not measured by a computer’s ability to spot trouble earlier, but by whether that insight successfully prevents compounding financial, social, and psychological damage.