GameScanner, the Mindway AI tool now used by lottery and betting operators across dozens of markets, is the clearest example of a growing category: player-protection software marketed as being built on “neuroscience.” The underlying brain science is real and well documented — particularly around near-miss outcomes and dopamine — but it describes how gambling affects the brain, not how an algorithm reads a player’s clickstream. The actual evidence for tools like GameScanner comes from a narrower, more mundane source: statistical models trained on account-level behavioral data, independently sample-tested by Gaming Laboratories International (GLI). Both strands matter, and conflating them is where most marketing claims overreach.
Key takeaways
- GameScanner’s accuracy claims were sample-tested by GLI in 2020, which found that at least 88.5% of manually assessed problem gamblers were classified as potential problem gamblers by the tool, and 99% as at-risk or problem gamblers.
- The neuroscience behind player-protection AI mostly traces to near-miss research: a 2010 study found gambling severity predicted a greater dopaminergic midbrain response to near-miss outcomes, though a later study found amplified brain responses without clear dopamine evidence.
- Most commercial AI player-protection tools don’t scan brains — they run machine-learning models on behavioral proxies like deposit frequency and account depletion, a method pioneered in peer-reviewed work by researchers Michael Auer and Mark Griffiths.
- Independent reviewers report false-positive rates for AI gambling-harm detection in the range of 15–30%, and the EU AI Act has banned AI systems that exploit psychological or economic vulnerabilities since February 2025.
- The same behavioral data that powers harm detection can power retention marketing, which is why GamblScout.com’s algorithm checks whether an operator’s “protection” stack is organizationally separate from its marketing stack.
Table of contents
What GameScanner actually is
GameScanner is a player protection software tool that allows gambling operators to better monitor their players, using algorithms to spot and flag problematic behavior in a player’s online gambling.
Mindway AI describes it as a software solution that helps gambling operators boost player protection through a combination of AI, neuroscientific insights and expert assessments.
It is built with input from a team of experts in neuroscience and problem gambling, including researchers, psychologists, who help the system spot risk factors before they become a habit.
Operators pitch this as a “virtual psychologist” running continuously in the background.
GameScanner is described as leveraging AI and expert human assessment to deliver real-time, individualised and full player-base detection of at-risk or problematic gambling behaviours, offering a “Virtual Psychologist” experience.
In practice, GameScanner has moved well past a single niche product.
The company reports its tools are used in 37 countries, helping over 7.7 million players monthly.
A notable land-based expansion came in September 2025, when the company partnered with Crown Resorts in Australia to launch a player-protection program across its casinos
, and on the digital side,
the technology has been integrated into sportsbook platforms to enable real-time analysis of player behavior and proactive detection of at-risk gaming patterns.
For context on how automated systems are assessed across the wider responsible-gambling stack, see our analysis of how AI detects problem gambling before the player even realizes it.
The neuroscience of near-misses and dopamine
The “neuroscience” claim in player-protection marketing usually points back to one well-replicated finding: the brain treats a near-miss almost like a win.
The brains of problem gamblers react more intensely to “near misses” than casual gamblers, possibly spurring them on to play more, according to research published in The Journal of Neuroscience.
The study detected strong responses in the midbrain, an area associated with addiction that is packed with dopamine-releasing brain cells.
A follow-up fMRI study sharpened this into a dose-response finding.
Across a group of regular gamblers, near-miss outcomes were associated with a significant response in the ventral striatum, and gambling severity, measured with the South Oaks Gambling Screen, predicted a greater response in the dopaminergic midbrain to near-miss outcomes.
The researchers concluded that near-miss events during gambling recruit reward-related brain circuitry in regular players.
This is where a skeptical reading matters: not every study agrees on the mechanism.
One later study reasoned that pathological gamblers might have amplified striatal sensitivity to near-misses, and found that near-misses elicited higher motivation to continue gambling and increased striatal responses compared with full-misses
— but
that same research found no evidence that these responses are influenced by dopamine specifically
. In other words, the “it’s dopamine” shorthand used in a lot of trade-press coverage overstates how settled the pharmacology actually is, even though the behavioral effect — near-misses keeping people playing — is well established. This neurological reality is also the scientific backbone behind our separate investigation into how unethical casinos manipulate UI to exploit loss-aversion triggers, since near-miss framing and loss-aversion design both lean on the same reward-circuitry research.
From brain scans to behavioral proxies
No commercial operator is putting players in an MRI scanner before a deposit. What GameScanner and comparable tools actually do is infer a near-miss-sensitive, loss-chasing brain state from account logs — a method with its own, separate research lineage built on player-tracking data rather than neuroimaging. The most cited body of this work comes from researchers Michael Auer and Mark Griffiths.
Their research has shown that self-reported problem gambling can be predicted by AI algorithms with high accuracy based on player tracking data.
Their operationalization of “chasing losses” — a core diagnostic marker for gambling disorder — is a good illustration of how abstract psychology gets turned into a queryable variable.
Using account-based tracking data from an online casino operator, they aimed to operationalize the behavioral indicator of loss chasing, one of the key harm markers in gambling disorder, analyzing records including bets, sessions, and monetary deposits.
They concluded that frequent session depositing reflected chasing losses better than other metrics tested, arguing that frequent depositing within a session or on a day and regular depletion of the gambling account may be indications of chasing and not being able to stop gambling.
In a separate, larger dataset,
the researchers were given access to the raw data of 1,287 players from a European online gambling casino who answered questions on the Problem Gambling Severity Index
, which let them validate behavioral proxies against a clinically recognized screening instrument rather than guesswork.
Account depletion is another recurring marker.
Studies have reported that problem gamblers tend to regularly deplete their accounts, spending all of their funds by the end of a gambling session.
And when different modeling approaches are compared,
random forest models have outperformed other approaches in identifying gambling-related harm among samples of online casino players.
This is the actual mechanical content behind phrases like “neuroscience-based AI”: pattern recognition across deposit timing, session length, and chasing frequency, calibrated against self-reported harm, not a live read of someone’s dopamine levels.
What the GLI validation actually shows
GameScanner is unusual in the sector because its performance claims were put to an independent lab rather than left as marketing copy.
The claims on the performance of GameScanner, which is based on artificial intelligence and neuroscience and aims to help igaming operators monitor behaviour of their customers, were tested based on a sample performed in August 2020 by Gaming Laboratories International.
| Claim tested | GLI finding |
|---|---|
| Classification of manually assessed problem gamblers as at-risk or potential problem gamblers | At least 99 per cent of subjects manually assessed as potential problem gamblers are classified as either at-risk or potential problem gamblers by GameScanner. |
| Classification of manually assessed problem gamblers specifically as potential problem gamblers | At least 88.5 per cent of subjects manually assessed as potential problem gamblers are classified as potential problem gamblers by GameScanner. |
Two caveats are worth flagging for an experienced audience. First, this was a sample test, not an ongoing independent audit cadence —
it validated a specific sample performed in August 2020
, and we have not found a published record of a repeat third-party test at the same rigor since. Second, GLI’s own terms of reference state that
its reports will only reflect GLI’s assessment of the conformity of test results to the applicable industry or regulatory standard or jurisdictional requirement specified in the proposal
— meaning the lab confirmed the stated numbers held up on the submitted sample, not that the tool generalizes identically across every jurisdiction, game type or demographic it’s since been deployed into. For a broader look at how third-party certification gets used (and sometimes oversold) across the industry, see our security, fraud detection and fair play coverage.
False positives, dual-use, and the EU AI Act
No behavioral-prediction model is free of misclassification, and vendors rarely publish their own false-positive rates. Independent industry reporting puts the scale of the problem in context:
current detection systems reportedly generate false positive rates between 15 and 30 percent depending on the sensitivity thresholds configured by operators, a range that creates substantial customer experience and commercial friction.
That is why most deployments keep a human in the loop rather than letting the model act alone —
AI systems can generate false positives, and players can always request human review of automated decisions.
A regulator-facing review of the field reached a similar conclusion about maturity.
Researchers found that in contexts where explainability is essential, current tools still show important limitations.
In at least one case examined, applying off-the-shelf technology to responsible-gambling settings showed limited value in distinguishing risk between products.
That gap between marketing confidence and operational nuance is precisely why GamblScout.com built a dedicated test for operator-side friction — see the friction check on deposit limits — because a harm-detection model is only protective if the intervention it triggers is easy for the player to act on.
There is also a structural conflict of interest worth naming directly. The behavioral features that flag chasing or depletion are often drawn from the same data warehouse that powers personalization and retention. One recent industry analysis put it bluntly:
in many operators, the same neural networks that identify a player in distress are drawn from the same data lakes, built by the same engineers, and are judged against many of the same key performance indicators as the systems that personalize marketing and maximize lifetime value.
The same piece argues that a genuinely protective deployment requires separation:
the marketing stack should not inform the safer-gambling stack.
Regulators are now writing some of this into law.
Since 2 February 2025, Article 5 of the EU AI Act has prohibited the placing on the market, putting into service or use of certain AI systems for manipulative, exploitative and social control or surveillance practices that inherently violate fundamental rights and Union values.
The vulnerability-exploitation provision is directly relevant to gambling platforms:
Article 5(1)(b) prohibits the production or use of AI systems that exploit any of the vulnerabilities persons due to their age, disability or a specific social or economic situation, with the objective, or the effect, of materially distorting the behaviour of that person in a manner that causes or is reasonably likely to cause significant harm.
The practical examples regulators have published include systems that
target older people with deceptive offers and scams to cause them financial harm
— a reminder that the line between a “protection” model and an “exploitation” model is the intent and incentive structure behind it, not the underlying math. UK readers will recognize the same tension from affordability rules; see our breakdown of affordability checks in the UK for how a parallel debate has played out around financial-vulnerability data.
Beyond GameScanner: the wider field
GameScanner is the most visible brand in this space, but it sits inside a longer research tradition of commercial behavioral-tracking tools, including earlier systems built on the same academic groundwork from Auer and Griffiths.
Auer and Griffiths have argued that good tools which track a player’s behavior should be able to support informed player choice, and also help online gambling operators gain more insight into their players’ behavioral patterns
— a framing that predates the current wave of AI branding by over a decade. What has changed is the modeling technique (gradient boosting and neural networks instead of simple thresholds) and the packaging (dashboards, “explainable AI” language, award submissions) rather than the core data signals themselves: deposits, session length, bet frequency, and chasing patterns.
The practical test for any of these systems isn’t whether it can name-drop neuroscience, but whether the intervention it triggers is enforced consistently once a player is flagged. We’ve examined this question directly in our review of automated cooling-off periods and which casinos enforce them best, which is the downstream step that determines whether a detection model like GameScanner produces a real-world safeguard or just a compliance log entry.
Frequently asked questions
What is GameScanner, exactly?
GameScanner is a player protection software tool built by Mindway AI that allows gambling operators to monitor players and uses algorithms to flag problematic behaviour
. It is not a diagnostic medical device; it is a behavioral-pattern classifier that operators use to prioritize which accounts get human review or an automated intervention.
Is the neuroscience claim behind these tools legitimate?
The underlying brain research on near-misses and reward circuitry is real and peer-reviewed, as documented in Journal of Neuroscience studies. However, commercial tools don’t measure brain activity directly — they infer risk from account behavior like deposit timing, a method validated separately through player-tracking research rather than neuroimaging.
How accurate has GameScanner been shown to be?
In a 2020 sample test,
GLI found at least 99% of manually assessed problem gamblers were classified as at-risk or potential problem gamblers, and at least 88.5% were classified specifically as potential problem gamblers.
That figure reflects one sample test, not a continuously re-audited benchmark.
Can AI player-protection tools misclassify healthy players?
Yes. Industry reporting puts false-positive rates for behavioral detection systems between roughly 15% and 30% depending on sensitivity settings, which is why most operators route flagged accounts through human review rather than fully automated restrictions.
Does any law restrict how this data can be used?
In the EU, Article 5 of the AI Act has banned AI systems that exploit age, disability, or economic vulnerabilities to materially distort behavior and cause harm, in force since February 2025. This affects both harm-detection tools and the marketing systems that share their underlying data.
Methodology
For topics in the AI player-protection category, GamblScout.com’s algorithm weighs whether an operator’s responsible-gambling claims are backed by third-party validation (such as GLI or regulator filings) rather than vendor press releases alone, whether behavioral-intervention thresholds and friction levels are independently testable, and whether public evidence suggests separation between an operator’s retention/marketing data pipeline and its harm-detection pipeline. These signals feed into the broader scoring approach described on our scoring system and algorithmic weights hub, and sit alongside the review philosophy outlined in our core principles and the problem with “human” reviews.
Gambling involves risk. Only play with money you can afford to lose and use the deposit limits and self-exclusion tools available in your jurisdiction.
