Can AI Detect Problem Gambling Before the Player Does?

An online casino potentially knows considerably more about a player's gambling behavior than the player remembers.

It can record when someone logs in, how long they play, how much they deposit, how frequently deposits occur, changes in stake size, losses, withdrawals, rejected payments, bonuses used and whether responsible-gambling controls have been activated.

Individually, these are transactions. Combined over thousands or millions of sessions, they become behavioral data.

Artificial intelligence is increasingly being applied to that data with a potentially transformative objective: identifying patterns associated with gambling harm before the player recognizes the problem themselves.

In 2026, this is moving from theory toward practical risk modeling. A newly published University of Sydney study found that machine-learning models could identify high-risk online bettors using only 30 days of behavioral data, achieving an area under the receiver operating characteristic curve [ AUROC] of approximately 0.74 - 0.75. Adding selected self-reported information increased model performance as high as 0.85.

That is promising. It is not, however, the same as giving an algorithm the ability to diagnose gambling addiction.

The distinction could become one of the most important responsible-gambling debates of 2027.

The Casino Has Something the Player Does Not: Perfect Transaction Memory

Human beings are imperfect observers of their own behavior.

Someone may remember depositing €100 but forget several €20 top-ups later that evening. A losing weekend can be rationalized against a memorable win months earlier. Time spent gambling may also feel considerably shorter than it actually was.

Research by Michael Auer and Professor Mark Griffiths comparing self-reported and actual online gambling expenditure found that players with greater losses tended to have greater difficulty accurately estimating their expenditure.

An algorithm does not have this memory problem.

Consider a hypothetical player's behavior:

None of these variables alone proves gambling harm. Together, however, they describe a rapidly changing behavioral trajectory.

That is where machine learning becomes potentially valuable.

What Does an AI Gambling-Risk Model Actually Look For?

There is no universal "problem gambling algorithm."

Models can be trained using combinations of account and behavioral variables and then compare emerging patterns against outcomes observed in known higher-risk groups.

The UK Gambling Commission already requires remote operators to monitor categories including customer spend, patterns of spending, time spent gambling, behavioral indicators, customer contact, use of gambling-management tools and account indicators.

Machine learning can go further by examining interactions between variables that would be difficult for a human compliance employee to continuously monitor.

A 2026 study by Robert Heirene and colleagues analyzed 1,470 customers of two Australian sports and race betting sites. The researchers found that important account-based predictors included age, deposits per active day, average stake and days since betting. Behavioral data alone produced useful classification performance, while adding self-reported measures - particularly the number of gambling accounts held and satisfaction with gambling - improved predictions.

This finding is particularly interesting because six months of behavioral history did not dramatically outperform the 30-day models.

In other words, potentially useful warning signals may emerge relatively quickly.

A Technical Expert's Warning: Prediction Still Needs Humans

Dr Michael Auer, a gambling-data researcher who has published extensively with Professor Mark Griffiths on AI, behavioral tracking and responsible gambling, argues that machine learning is already powerful enough to identify risk - but should not operate without human involvement.

Speaking to SBC News about AI-powered player protection, Auer said:

"Automated processes alone cannot deter problem gambling at its source."

He argued that personalized automated interventions can be useful, but trained professionals may still need to contact players and assess their behavior directly.

That distinction matters.

The best 2027 system may therefore not be AI instead of human responsible-gambling teams.

It may be AI finding the signal, followed by humans deciding what the signal means and what proportionate intervention is appropriate.

Why 2026 Is an Important Turning Point for AI and Problem Gambling Solutions?

Before the current generation of predictive systems, responsible-gambling controls were largely reactive.

A player exceeded a threshold. A payment failed. A customer contacted support. A deposit limit was activated. Someone self-excluded.

The emerging model is predictive. Instead of focusing on whether a customer has a gambling issue,  a system can ask, "Is this customer's behavior moving toward a pattern historically associated with elevated risk?"

The UK regulatory framework is already moving in this direction conceptually. Gambling Commission guidance requires systems to flag indicators of potential harm in a timely manner and states that automated processes may be appropriate where customer volumes make manual monitoring impractical. Strong indicators must trigger timely automated action, while affected customers must have an opportunity to contest automated decisions.

The technology is therefore developing alongside regulation rather than in isolation.

Why Getting This Right Matters: The Scale of Gambling Harm

The potential public-health value is substantial.

The World Health Organization, as cited by us regularly, estimates that approximately 1.2% of the world's adult population has a gambling disorder. WHO also cites estimates suggesting that people gambling at harmful levels account for around 60% of gambling losses - or industry revenue.

Great Britain's newest statistics demonstrate the scale at a national level.

The Gambling Commission's 2025 annual survey, published in July 2026, estimates that 2.4% of adults scored eight or above on the Problem Gambling Severity Index (PGSI), equivalent to approximately 1.3 million adults in Great Britain. Among people who had gambled during the previous 12 months, 4.0% scored eight or higher.

There is also an important gender dimension.

Among people gambling during the previous year, 5.3% of male participants scored PGSI 8+, compared with 2.7% of female participants. However, lower prevalence among women should not be interpreted as low importance: 11.4% of female gamblers scored between 1 and 2, indicating a much broader population showing some risk indicators.

For AI models, gender should therefore be treated carefully. A system trained predominantly on historically male high-risk populations could theoretically become less sensitive to different patterns of harm among women. That makes representative training data and model validation increasingly important.

The Problem AI Creates While Solving Another

Predictive player protection creates an uncomfortable contradiction.

The same casino data that can identify vulnerability can also be used to increase engagement.

An algorithm might discover that a player responds strongly to free spins after losses, prefers playing after 11 p.m. and is more likely to deposit after receiving a push notification.

One AI system could interpret those signals as marketing opportunities.

Another could interpret them as warning signs.

That makes the objective of the algorithm critically important.

There are also privacy and fairness questions. In Europe, automated profiling sits within the wider GDPR framework, and the European Data Protection Board has specific guidance covering automated individual decision-making and profiling.

A false negative may leave a vulnerable customer unprotected. But a false positive could incorrectly label an ordinary player as high risk, restrict an account, or produce intrusive interventions.

Accuracy alone is therefore insufficient.

PESTEL Snapshot: AI Player Protection in 2027

Geography Will Shape How Fast AI Protection Develops

AI player protection will not evolve uniformly.

Britain already has detailed remote customer-interaction requirements, including monitoring from account opening and automated responses to strong indicators of harm.

Australia presents a different regulatory environment. Online casino services are prohibited under the Interactive Gambling Act, while licensed online wagering operates within a tightly controlled framework. Interestingly, Australia's communications regulator published a dedicated report in April 2026 examining AI applications across interactive gambling and other regulated sectors, demonstrating that regulators themselves are now actively studying the technology.

European operators must additionally consider GDPR requirements surrounding profiling and personal data.

For international casino groups, a single global responsible-gambling algorithm may consequently be less realistic than locally validated models operating within jurisdiction-specific rules.

2027 Forecast: From Responsible Gambling Tools to Responsible Gambling Systems

CasinoAppReview expects the next development to be a shift from isolated tools toward integrated risk engines.

Today's casino might offer deposit limits, reality checks, and self-exclusion as separate features.

Tomorrow's system could continuously calculate a changing risk score based on behavioral signals.

A low-risk player might simply receive spending information.

Increasing risk could trigger personalized feedback or encourage a deposit limit.

Higher-risk patterns might suppress bonuses and marketing.

Strong indicators could trigger human review, temporary restrictions or direct intervention.

The important development is not merely automation. It is graduated intervention.

The UK Gambling Commission's framework already reflects this philosophy: identify, act and evaluate, with stronger actions where earlier interventions have failed.

So, Can AI Know Before the Player Does?

Potentially, yes, but with an important qualification.

AI cannot look inside someone's mind and determine whether they have a gambling disorder. Nor should a casino algorithm be treated as a clinical diagnosis.

What it can do increasingly well is recognize behavioral change.

It can notice that deposits are becoming more frequent. Stakes are rising. Sessions are becoming longer. Gambling is occurring more often. Previous limits are being changed. The player's behavior is moving away from their historical baseline.

And it can remember every transaction when the player cannot.

That may be AI's most important contribution to responsible gambling.

The casino industry has spent years applying algorithms to understand what players want to play next.

The more consequential question for 2027 may be whether the same computational power can determine when they should stop.

References / Works Cited

Australian Communications and Media Authority. (2026). AI in telecommunications, media and interactive gambling. ACMA AI Report

Auer, M., & Griffiths, M. D. (2017). Self-reported losses versus actual losses in online gambling: An empirical study. Journal of Gambling Studies, 33, 795 - 806. PubMed record

Auer, M., & Griffiths, M. D. (2023). Using artificial intelligence algorithms to predict self-reported problem gambling with account-based player data in an online casino setting. Journal of Gambling Studies, 39(3), 1273 - 1294. PubMed study

European Data Protection Board. (2018). Guidelines on automated individual decision-making and profiling. EDPB guidance

Gambling Commission. (2023). Customer interaction guidance for remote gambling licensees: Formal guidance under SR Code Provision 3.4.3. UK Gambling Commission guidance

Gambling Commission. (2026). Gambling Survey for Great Britain: Annual report 2025 - Official statistics. GSGB 2025 statistics

Heirene, R. M., Zhang, E., Vanichkina, D., de Leau, C. T., Huynh, E. L. Y., & Gainsbury, S. M. (2026). Predicting problem gambling among online sports and race bettors: Assessing the value of machine learning using behavioural and self-reported data. Journal of Behavioral Addictions, 15(2), 846 - 862. PubMed research paper

SBC News. (2024). Dr Michael Auer: Neccton is here to integrate safe play into the player's experience. Dr Michael Auer interview

World Health Organization. (2024). Gambling. WHO gambling fact sheet