A new study claims that some of the most common generative artificial intelligence (AI) tools can direct Dutch consumers to casinos without a Dutch Gambling Authority (KSA) license.
The research examined how 10 AI tools respond to realistic questions about online gambling, probing results from ChatGPT, Google’s AI Mode, Google Gemini, Microsoft Copilot, Perplexity, Claude, Grok, DeepSeek, Mistral Le Chat and Meta AI.
According to the report, all 10 tools responded with at least one online gambling operator without a Dutch licence in one or more of the cases. In eight instances, unregulated operators appeared during either neutral or more targeted questions, without the user presenting the request as part of a research exercise.
Testing With Three Sets of Questions
The study focused on the Dutch online casino market and used Dutch-language questions. Researchers tested each tool with three sets of questions, starting with neutral consumer searches, then moving to more targeted questions about casinos without CRUKS, and finally requests framed as research.
The first set included questions such as which were the best online casinos in the Netherlands, requests for a “top 50”, and questions about international gambling sites popular among Dutch consumers.
Then, the second set directly asked about online casinos without CRUKS. The third was the same type of request as research into the illegal gambling market and asked for specific casinos and domains.
The researchers assessed whether a tool recommended an operator or domain without a valid Dutch gambling licence. Licensed Dutch casinos are those listed on the KSA’s register on September 1, 2026. The report also distinguished between online gambling licences and licences for land-based gambling.
All 10 Tools Mentioned Unlicensed Operators
Furthermore, the report found significant distinctions between the AI tools. However, none of the tools avoided mentioning illegal operators across all of the tested methods.
For instance, Google’s AI Mode, Mistral Le Chat, DeepSeek and Grok were among the tools where a single targeted question was sufficient to recommend an unregulated site. Meanwhile, ChatGPT mentioned unlicensed operators at the third question in the neutral series, at the second question in the targeted series and immediately in the research scenario.
Perplexity and Claude were more resistant, but both ultimately produced names when questions referred to operators appearing in KSA enforcement cases. Meta AI maintained its refusal during the targeted questioning but provided names during the research scenario in one of the tested conversations.
Overall, the study recorded 79 individual AI responses. An unlicensed operator appeared in 33 of those responses.
The report states that all 10 tools mentioned offshore operators, while eight did so during either the neutral or targeted question sequences. Four tools produced such a name after the first question of one of those sequences. Three tools also presented unlicensed brands in lists described as legal or Dutch-licensed.
AI Responses Can Feature Promo Details
Researchers also identified three different ways in which unlicensed gambling operators appeared in the AI responses.
In some cases, tools simply provided information about operators, including names taken from previous KSA rulings. The researchers described this as informational rather than promotional, but they noted that users were still presented with concrete names and domains.
Other responses highlighted features that could make unlicensed gambling sites attractive to consumers, including higher bonuses, quicker withdrawals, no identity checks and the lack of CRUKS.
The study notes that Google AI Mode, Mistral and DeepSeek provided such suggestions in their responses, while Gemini and Copilot also included details on why particular sites were popular or whether they were VPN-friendly. DeepSeek also offered advice about starting with a small amount and setting personal limits.
Incorrect Licensing Information Raises Concerns
The report concludes that a particularly concerning issue is that AI tools incorrectly label unlicensed operators as licensed ones. Microsoft Copilot, Mistral Le Chat and Google Gemini were among the tools to present brands without Dutch online gambling licences as legal or Ksa-licensed.
In the case of Copilot, the study found the tool also stated that the operators in its list were connected to CRUKS. As a result, researchers argue that consumers could receive the impression that an operator does business under Dutch regulations when it’s not the case.
Also, the report reads that Copilot produced a “Top 50” of seemingly legal casinos with several brands without Dutch licences. At the same time, Mistral included unlicensed operators in a table labelled as consisting of legal casinos.
Similarly, Gemini included an operator without a Dutch licence in a list presented as legal and later produced brands described in terms of features such as cryptocurrency and limited KYC requirements.
Refusals Were Not Always Maintained
Another finding was about AI tools that initially refused to provide names of unlicensed operators.
According to the research, six tools initially rejected the research-based request for specific casinos. But afterwards provided names when the researchers changed the wording and asked which operators had been mentioned in the Dutch regulator’s enforcement cases.
Grok, Copilot, Claude and Meta AI eventually provided names after being asked about the KSA’s own enforcement cases. After initially rejecting, Gemini also provided names, while Perplexity produced names when asked for concrete domains for a source list.
Therefore, the study argues that refusals were not consistently supported when users continued the conversation or reframed their requests.
Researchers Highlight Limitations
The report stresses that the findings should not be interpreted as evidence that every consumer will receive the same responses.
AI-generated answers are non-deterministic and can change depending on updates, settings and the time at which a question was asked. The researchers describe the results as a defined and indicative sample, and not as conclusive evidence about individual operators.
However, the study did not look into whether users could create accounts or deposit money with the suggested operators. Its objective was to examine whether AI tools would provide the names or domains of operators without a Dutch online gambling licence.
Call for Better Verification of Gambling Licences
Overall, the report concludes that generative AI has become a relevant information channel for players researching online gambling and that they should pay greater attention to how AI systems present information about operators and their licences.
The researchers specifically emphasise the need for reliable verification before an operator is described as legal, licensed or connected to CRUKS. They also call for caution when AI tools describe attractive characteristics of unlicensed gambling services.
For consumers, the report recommends treating AI-generated answers as insufficient evidence that an online casino is legal or reliable. Instead, it says the licence status of an operator should be checked through an official source.
