Dutch affiliate watchdog Keurmerk Verantwoorde Affiliates (KVA) has published research showing that popular generative AI tools can direct consumers to illegal online gambling operators — and in some cases present unlicensed sites as if they were legally authorised.
Scope of the findings
The KVA study, published at the end of September, tested ten widely used AI systems — including OpenAI's ChatGPT, Google Gemini, Microsoft Copilot and Mistral Le Chat — with a set of consumer-style questions about online casinos. According to the report, all ten systems named operators that do not hold a Dutch licence when asked relatively simple queries; eight did so even for neutral or lightly guided questions.
In several cases the tools went further: Google Gemini, Microsoft Copilot and Mistral Le Chat are reported to have presented unlicensed operators as legal or as licensed by the Dutch regulator (Kansspelautoriteit). ChatGPT was also found to recommend unlicensed providers across all three tested question series.
Why this matters
- KVA says consumers relying on AI answers may get a misleading impression of the legal status of gambling sites — a particular risk in a regulated sector such as online gambling.
- The study logged 79 recorded AI responses and is based on a bounded sample; KVA frames the results as indicative rather than exhaustive.
- KVA highlights the risk for vulnerable groups, including minors, who could be exposed to illegal gambling offers if they follow AI-generated recommendations.
Calls for verification and better safeguards
Steven Vrolijk of KVA urged AI developers to improve how systems handle information about licences and regulated providers. KVA recommends that developers verify the legal status of online gambling sites against authoritative sources, such as the Kansspelautoriteit licence register, before presenting them to users.
The Dutch findings are raised against the background of growing scrutiny of AI safety internationally — including recent lawsuits in several U.S. states targeting OpenAI over alleged safety shortcomings in ChatGPT — and add to broader debates about the reliability of generative models for consumer-facing advice.
Limitations and next steps
KVA characterises the report as based on a limited sample and calls on AI firms to take practical steps to reduce the risk of producing misleading or unlawful recommendations. The full research and supporting documents are linked in the organisation's release.
For regulators and platform developers the study underlines a recurring challenge: ensuring that AI systems consult and surface up-to-date, authoritative information when the legal status of services matters to user safety.
Source: Original press release
